An adaptive calibration method and system for pressure transmitters

Abstract

The present application relates to the technical field of pressure transmitter, and particularly relates to a self-adaptive calibration method and system for pressure transmitter. The method comprises the following steps: S1: collecting multiple groups of measurement sample data of a target pressure transmitter in a current time window; S2: constructing and training a measurement sample data encoder of a corresponding mode, encoding the measurement sample data as environment features of the corresponding mode, and extracting common features of different mode environment features; S3: taking the common features as a condition, performing multi-scale reserved noise addition on the environment features of the corresponding mode, training an error prediction network in a corresponding mode conditional diffusion model using the environment features of the corresponding mode after noise addition, and the error prediction network has the ability to fit the conditional environment feature distribution. Through the latest diffusion model reforming idea, the multi-scale reserved noise addition and the conditional sampling mechanism are used to improve the generalization ability of the model to unseen or rare working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pressure transmitters, in particular to a self-adaptive calibration method and system for pressure transmitters. BACKGROUND

[0002] As one of the key sensors in the fields of industrial automation, petrochemical industry, aerospace, transportation and environmental monitoring, pressure transmitters bear the task of converting the measured pressure signal into standardized electrical signal and reporting to the control system. In recent years, with the increasing complexity of process and harshness of on-site environment, the problems of measurement deviation, zero drift and sensitivity change of pressure transmitters in long-term operation are increasingly prominent, which directly affects the production safety and control accuracy. Since the on-site environment of pressure transmitters is complex and changeable, it is usually difficult to obtain high-quality labeled data covering all working conditions, so that the calibration model based on supervised learning performs poorly when migrating to a new scene, and ignores the fact that some high-performance calibration algorithms are difficult to run in real time on edge devices with limited computing resources, affecting the feasibility of online adaptive compensation. SUMMARY

[0003] In order to overcome the shortcomings of lack of data for unobserved or rare working conditions, the present application provides a self-adaptive calibration method and system for pressure transmitters.

[0004] The technical solution of the present application is: a self-adaptive calibration method for pressure transmitters, comprising the following steps:

[0005] S1: collecting multiple groups of measurement sample data of the target pressure transmitter within the current time window;

[0006] S2: constructing and training a measurement sample data encoder corresponding to the modal, encoding the measurement sample data as the environmental features of the corresponding modal, and extracting the common features of different modal environmental features;

[0007] S3: using the common features as conditions, performing multi-scale reserved noise addition on the environmental features of the corresponding modal, training the error prediction network in the corresponding modal conditional diffusion model using the environmental features of the corresponding modal after noise addition, and the error prediction network has the ability to fit the conditional environmental feature distribution;

[0008] S4: performing multi-scale reserved noise addition on the environmental features of the corresponding modal, using the error prediction network of the corresponding modal to sample, and generating the conditional environmental features of the corresponding modal;

[0009] S5: constructing and training a pressure transmitter calibration network, sending the conditional environmental features of different modes and the pressure data detected by the pressure transmitter within a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data.

[0010] Preferably, the collecting the multiple sets of measurement sample data of the target pressure transmitter within the current time window comprises: acquiring the multiple sets of measurement sample data of the target pressure transmitter within the current time window, the measurement sample data comprising instantaneous pressure data, ambient temperature data, temperature gradient data, installation posture data, and historical compensation data.

[0011] Preferably, the constructing and training the measurement sample data encoder of the corresponding modal to encode the measurement sample data as the environmental feature of the corresponding modal and extracting the common feature of the environmental features of different modalities comprises: constructing a measurement sample time sequence according to the measurement sample data in the time dimension, the measurement sample time sequence comprising an instantaneous pressure time sequence, an ambient temperature time sequence, a temperature gradient time sequence, an installation posture time sequence, and a historical compensation time sequence, and using a Transformer to extract time sequence features from the measurement sample time sequence, the time sequence features being the environmental features of different modalities.

[0012] Preferably, the constructing and training the measurement sample data encoder of the corresponding modal to encode the measurement sample data as the environmental feature of the corresponding modal and extracting the common feature of the environmental features of different modalities comprises: taking the average of the environmental features of different modalities as the initial common feature , is considered as The weighted coefficient of the weighted average is updated in the following manner:

[0013] ;

[0014] wherein, is the weighted coefficient of the environmental feature of the i-th modality; is the environmental feature of the i-th modality; is the environmental feature of the i-th modality; is the environmental feature of the i-th modality; is the set containing all the environmental features of the modalities; is a temperature parameter for controlling the smoothness of the weight; is the cosine similarity between two vectors; is the common feature at the current time; The common feature is generated in the following manner through the weighted coefficient:

[0015]

[0016] ;

[0017] ;

[0018] wherein, ​​​is a weighted mean value; is a weight adjustment coefficient; is a common feature at the next moment;

[0019] The final common feature is obtained through iteration .

[0020] Preferably, the common feature is used as a condition to perform multi-scale reserved noise addition on the environment feature of the corresponding modal, an error prediction network in the conditional diffusion model of the corresponding modal is trained using the environment feature after noise addition of the corresponding modal, the error prediction network has the ability to fit the conditional environment feature distribution, and the conditional environment feature distribution is close to the common feature, and the method comprises the following steps: regarding the common feature as a condition for the variance prediction network in the diffusion model; and the multi-scale reserved noise addition process is described as follows:

[0021] ;

[0022] wherein, is the environment feature after noise addition when ; is the environment feature without noise addition; is random noise, ; is an adjustment coefficient of the noise scale, used to control the noise scale;

[0023] The variance prediction network of the diffusion model of the corresponding modal is trained using the environment feature after noise addition of different modalities and different noise scales, respectively, to fit the conditional environment feature distribution of the corresponding modal, and the error prediction network is described as , is a training parameter of the network.

[0024] Preferably, the environment feature of the corresponding modal is subjected to multi-scale reserved noise addition, and the error prediction network of the corresponding modal is used for sampling to generate a conditional environment feature of the corresponding modal, and the method comprises the following steps: the environment feature of the corresponding modal is subjected to noise addition according to the multi-scale reserved noise addition process to obtain a noise-added environment feature of the corresponding modal , and the conditional environment feature of the corresponding modal is generated by sampling in the following manner :

[0025] ;

[0026] ;

[0027] wherein, is the differential of ; is the first derivative of ; Depend on the step size to control the noise level, and ; is the time step of the sampling; is the score function of ; is the differential of the time step ; is the interference term satisfying the It process; and is the .

[0028] Preferably, the pressure transmitter calibration network is constructed and trained, and the conditional environment features of different modalities and the pressure data detected by the pressure transmitter in a preset time period are input into the pressure transmitter calibration network to predict and calibrate the pressure data, comprising: constructing and training a pressure transmitter calibration network, using self-Attention to aggregate conditional environment features of different modalities, calling the aggregated conditional environment features mixed conditional environment features, constructing a time sequence according to the time dimension of the pressure data detected by the pressure transmitter in a preset time period, using Transformer to extract the time sequence features of the pressure data, inputting the mixed conditional environment features and the time sequence features of the pressure data into MLP, predicting the pressure data at the current time and selecting a compensation strategy.

[0029] Preferably, the mixed conditional environment features and the time sequence features of the pressure data are input into the MLP, the pressure data at the current time is predicted, and a compensation strategy is selected, comprising: according to the deviation degree of the predicted pressure data relative to its historical sequence, evaluating the confidence of the predicted pressure data at the current time, when the confidence is lower than a first preset threshold, weighting and fusing the predicted pressure data and the original pressure measurement data to output a final calibration value; when the confidence is lower than a second preset threshold, a low confidence alarm is issued, and the second preset threshold is smaller than the first preset threshold.

[0030] Preferably, the mixed conditional environment features and the time sequence features of the pressure data are input into the MLP, the pressure data at the current time is predicted, and a compensation strategy is selected, comprising: when the predicted pressure data is greater than or equal to a first preset threshold, selecting a current optimal compensation strategy from a plurality of preset compensation strategies, or dynamically adjusting the parameters of the current compensation strategy; the plurality of compensation strategies are optimized for different pressure ranges, environmental change intensities, or device operating states.

[0031] Preferably, an adaptive calibration system for a pressure transmitter comprises:

[0032] A data acquisition module is configured to collect a plurality of sets of measurement sample data of a target pressure transmitter in a current time window.

[0033] a common feature extraction module configured to construct and train a measurement sample data encoder of a corresponding modality, encode the measurement sample data as an environmental feature of the corresponding modality, and extract a common feature of different modality environmental features;

[0034] a network training module configured to use the common feature as a condition, perform multi-scale reserved noise addition on the environmental feature of the corresponding modality, and train an error prediction network in the corresponding modality conditional diffusion model using the noise-added environmental feature of the corresponding modality, the error prediction network having the ability to fit the conditional environmental feature distribution;

[0035] a feature generation module configured to perform multi-scale reserved noise addition on the environmental feature of the corresponding modality, sample using the error prediction network of the corresponding modality, and generate a conditional environmental feature of the corresponding modality;

[0036] a prediction calibration module configured to construct and train a pressure transmitter calibration network, and input the conditional environmental features of different modalities and pressure data detected by a pressure transmitter in a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data.

[0037] The present application has the following advantages:

[0038] 1. The present application can generate conditional environmental features with similar distribution to the real distribution for training or online completion by modeling and sampling the environmental features in the latent space using the conditional diffusion model, which means that the calibration network can be made more experienced when encountering rare temperature gradients, special installation attitudes or transient disturbances, thereby reducing the need for artificial field calibration and frequent re-calibration, saving maintenance costs and improving the trust of continuous operation of the equipment.

[0039] 2. The present application uses reserved noise addition rather than completely random generation, which preserves large-scale physical information and time sequence continuity in the original measurement during the sampling process, avoids generating abnormal situations that violate physical laws, and thus ensures the credibility and consistency of the conditional environmental features used for training or online completion in the physical sense.

[0040] 3. The present application can automatically reduce the influence of abnormal or noise modalities during the construction of common features by iteratively updating the common features based on cosine similarity and setting an iterative convergence criterion, which prevents abnormal samples from polluting the global environmental description, thereby ensuring the adaptability and robustness of the calibration model in long-term operation. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of the adaptive calibration method for pressure transmitters according to the present application.

[0042] Figure 2 Structure diagram of adaptive calibration system for pressure transmitter of the present application. DETAILED DESCRIPTION

[0043] The present application will now be described more fully hereinafter with reference to the accompanying drawings, in which currently preferred embodiments of the application are shown. The application may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided by way of thorough and complete disclosure of the application and are fully presented for purposes of full and enabling disclosure of the application to those skilled in the art.

[0044] Embodiment 1: An adaptive calibration method for a pressure transmitter, as shown in Figure 1 comprising the following steps:

[0045] S1: Collecting multiple sets of measurement sample data of the target pressure transmitter within a current time window;

[0046] Obtaining multiple sets of measurement sample data of the target pressure transmitter within a current time window, the measurement sample data including instantaneous pressure data, ambient temperature data, temperature gradient data, installation attitude data and historical compensation data.

[0047] It needs to be explained that the measurement sample data includes the following:

[0048] Instantaneous pressure data: original sequence of instantaneous pressure values recorded by the target pressure transmitter at each sampling time; Ambient temperature data: sampling sequence of the temperature sensor at the installation location or the adjacent environment; Temperature gradient data: gradient sequence obtained by a multi-point temperature sensor or through adjacent sensor temperature difference calculation; Installation attitude data: attitude time sequence measured by the three-axis acceleration / angular velocity sensor of the pressure transmitter installation angle, orientation or; Historical compensation data: compensation parameter sequence of the pressure transmitter historical calibration / compensation coefficient, last calibration record or updated during operation.

[0049] S2: Constructing and training a measurement sample data encoder of the corresponding modal, encoding the measurement sample data as the environmental features of the corresponding modal, and extracting the common features of different modal environmental features;

[0050] According to the measurement sample data, a measurement sample time sequence is formed according to the time dimension, the measurement sample time sequence includes an instantaneous pressure time sequence, an ambient temperature time sequence, a temperature gradient time sequence, an installation attitude time sequence and a historical compensation time sequence, a time sequence feature is extracted from the measurement sample time sequence using a Transformer, and the time sequence feature is an environmental feature of different modal.

[0051] It needs to be explained that the original measurement sample data is uniformly preprocessed:

[0052] Time alignment: interpolate or downsample each modal data according to a unified time reference, to ensure the formation of a synchronous measurement sample time sequence within the current time window;

[0053] Filtering and denoising: band-pass or low-pass filtering is used on the instantaneous pressure and temperature sequence to remove quantization noise and power frequency interference;

[0054] Standardization: normalize each mode according to historical mean and standard deviation or according to a sliding window, so that different modal value scales have the ability to directly input subsequent encoders;

[0055] Missing value processing: linear or spline interpolation is used for short-time missing data, and if the missing value exceeds the threshold, mark the time window as defective data and trigger the re-sampling or rollback strategy;

[0056] The obtained synchronized modal data is constructed into a measurement sample time sequence according to the time dimension, obtaining the instantaneous pressure time sequence, ambient temperature time sequence, temperature gradient time sequence, installation attitude time sequence and historical compensation time sequence. Through the Transformer, the fixed-dimensional time sequence feature vector is generated by pooling on the time dimension (such as taking the last time step output or using self-attention weighted pooling) or global average pooling. This vector is the environmental feature of the corresponding mode.

[0057] Take the average of different modal environmental features as the initial common feature , as , Update the weighted coefficients of the weighted average in the following way:

[0058] ;

[0059] Where, is the weighted coefficient of the th modal environmental feature; is the environmental feature of the th mode; is the environmental feature of the th mode; is a set containing all modal environmental features; is a temperature parameter used to control the smoothness of the weight; is the cosine similarity between two vectors; is the common feature at the current time;

[0060] Generate in the following way through the weighted coefficient:

[0061] ;

[0062] ;

[0063] Where, is the weighted mean; is the weight adjustment coefficient; The common characteristics of the next moment;

[0064] Obtain the final common features through iteration .

[0065] It should be explained that for each modality such as instantaneous pressure, ambient temperature, temperature gradient, installation posture and history compensation, a pre-trained measurement sample data encoder (such as a time series encoder based on Transformer) is used to obtain the modal environment feature vector , is the vector dimension; for each Perform standardization: first calculate the mean of each dimension in the training set and standard deviation and use Do normalization; after normalization, Normalize for cosine similarity calculation,

[0066] Calculate the arithmetic mean of all modal environment features as the initial common feature: ;

[0067] in is a set containing all modes, is the modal number, Assigned to , as the initial value of iteration;

[0068] For each mode , calculate the cosine similarity, where is the temperature parameter ( ), used to control the smoothness of the weights; The smaller, The sharper it is, the easier it is to amplify the most similar mode; The larger it is, the smoother the weights are; is the weight adjustment coefficient, the larger Tends to preserve historical commonalities to enhance smoothness, small Enhance responsiveness to changes in the current mode;

[0069] For short-term missing modes (intermittent sensor loss), when calculating the initial With weight Remove the missing modal from the collection Temporarily remove and renormalize the weights with the actual existing mode;

[0070] For abnormal modes (judged to be sensor failure or severe noise), they are marked as abnormal through anomaly detection before cosine similarity calculation, and former general Set to the minimum value or directly set the corresponding Set to , thereby inhibiting its impact.

[0071] If continuous All time window modes are marked as abnormal, which should trigger a maintenance / calibration prompt and permanently exclude the mode until manual intervention;

[0072] Through cosine similarity and temperature parameter of The weighted approach adaptively distributes the contribution of different modalities to common features. It can automatically weaken the influence of abnormal modalities when multimodal features are inconsistent or there are abnormal readings, and enhance the representativeness of common features.

[0073] pass The smooth update mechanism balances the trade-off between historical commonalities and current observations, enhancing the robustness of common features to short-term disturbances while ensuring adaptability to long-term environmental changes.

[0074] S3: Using the common features as conditions, performing multi-scale preservation and noise addition on the environmental features of the corresponding modality, and using the noise-added environmental features of the corresponding modality to train an error prediction network in a conditional diffusion model of the corresponding modality, wherein the error prediction network has the ability to fit the distribution of conditional environmental features;

[0075] The common features Considered as a condition of the variance prediction network in the diffusion model, the multi-scale preserving noise addition process is described as follows:

[0076] ;

[0077] in, for Environmental characteristics after adding noise; is the environmental feature without noise; is random noise, ; is the adjustment coefficient of the noise scale, which is used to control the noise scale;

[0078] The variance prediction network of the corresponding modal diffusion model is trained using the environmental features after adding noise of different modes and noise scales, and the conditional environmental feature distribution of the corresponding mode is fitted. The error prediction network is described as , The training parameters of the network.

[0079] It should be noted that, when close to , closer to (keep the proportion of the original content high); when close to 0, closer to pure Gaussian noise (disturbance strong); to enhance the multi-scale property, each sample on the training set is randomly sampled (uniformly or according to a preset distribution) to obtain a multi-scale set of noisy samples;

[0080] The error prediction network aims to recover and predict the noise component from the noisy sample and the common condition and the time step , so the following mean square error loss is used:

[0081] ;

[0082] During training, each is trained separately according to the mode (that is, an independent error prediction network is used for each mode), a multi-layer MLP is used as the backbone of , and the input of the network is the concatenated vector; among them, the condition access method uses direct concatenation and sends it into the network; layer normalization, residual connection and dropout are used in the network, and gradient clipping is used during training to prevent gradient explosion.

[0083] S4: The environment features of the corresponding mode are multi-scale reserved and added with noise, the error prediction network of the corresponding mode is used for sampling to generate the conditional environment features of the corresponding mode;

[0084] The environment features of the corresponding mode are added with noise according to the multi-scale reserved noise adding process to obtain the noisy environment features of the corresponding mode , and the conditional environment features of the corresponding mode are generated by sampling in the following way :

[0085] ;

[0086] ;

[0087] In the formula, is the differential of ; is the first derivative of ; depends on the step size to control the noise level, and ; is the time step of the sampling; is the score function of is the differential of the time step is the interference term satisfying the Itô process; and .

[0088] It needs to be explained that the conditional environment features generated by the above process are reconstructed by the common features, so that the distribution of the conditional environment features is close to the common features, thereby improving the interference caused by the missing or measurement error of single modal data on the subsequent network prediction value.

[0089] S5: Construct and train the pressure transmitter calibration network, and input the conditional environment features of different modalities and the pressure data detected by the pressure transmitter in the preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data.

[0090] The pressure transmitter calibration network is constructed and trained, the self-Attention is used to aggregate the conditional environment features of different modalities, the aggregated conditional environment features are called mixed conditional environment features, the pressure data detected by the pressure transmitter in the preset time period is constructed into a time sequence according to the time dimension, the time sequence features of the pressure data are extracted using the Transformer, the mixed conditional environment features and the time sequence features of the pressure data are input into the MLP, and the pressure data at the current time is predicted and the compensation strategy is selected.

[0091] According to the deviation degree of the predicted pressure data relative to its historical sequence, the confidence of the predicted pressure data at the current time is evaluated, when the confidence is lower than a first preset threshold, the predicted pressure data and the original pressure measurement data are weighted and fused to output a final calibration value; when the confidence is lower than a second preset threshold, a low confidence alarm is issued, and the second preset threshold is smaller than the first preset threshold.

[0092] ​​​It needs to be explained that the confidence of the predicted pressure data output by the MLP after the time sequence characteristics of the pressure data obtained by the Transformer from the mixed condition environment characteristics and the pressure measurement time sequence is quantified and weighted by three types of uncertainty: one is the deviation of the predicted value relative to the historical sequence (mapped to the confidence by z-score), the second is the model prediction uncertainty (estimated by the output variance or model ensemble / MC-Dropout), and the third is the consistency degree of the conditional environment (a consistency measure of the mixed condition and the training condition distribution). When the fused confidence is lower than the first preset threshold, the predicted value and the original pressure measurement value are linearly weighted and fused by the confidence normalization coefficient to output the final calibration value; when the confidence is further lower than the second preset threshold, a low confidence alarm is issued and a fallback or manual intervention process is adopted. When the predicted value meets the preset trigger condition (for example, the predicted pressure is greater than or equal to the preset pressure threshold or the deviation threshold), the current optimal compensation strategy is selected from the pre-set multiple compensation strategies based on historical revenue / cost / risk score or the compensation strategy parameters are dynamically adjusted based on the confidence, taking into account the calibration accuracy, execution cost and equipment safety.

[0093] When the predicted pressure data is greater than or equal to the first preset threshold, the current optimal compensation strategy is selected from the pre-set multiple compensation strategies, or the parameters of the current compensation strategy are dynamically adjusted; the multiple compensation strategies are optimized for different pressure ranges, environmental change severity or equipment operating state.

[0094] Embodiment 2: On the basis of embodiment 1, an adaptive calibration system for a pressure transmitter, as shown in Figure 2 , comprising:

[0095] A data acquisition module for collecting multiple sets of measurement sample data of the target pressure transmitter within the current time window;

[0096] A common feature extraction module for constructing and training a measurement sample data encoder corresponding to the modal, encoding the measurement sample data as the environmental characteristics of the corresponding modal, and extracting the common features of different modal environmental characteristics;

[0097] A network training module for using the common features as conditions and adding noise to the environmental characteristics of the corresponding modal in multiple scales, training the error prediction network in the corresponding modal condition diffusion model using the environmental characteristics of the corresponding modal after adding noise, and the error prediction network has the ability to fit the conditional environmental characteristic distribution;

[0098] A feature generation module for adding noise to the environmental characteristics of the corresponding modal in multiple scales, sampling using the error prediction network of the corresponding modal, and generating the conditional environmental characteristics of the corresponding modal;

[0099] A prediction calibration module is configured to construct and train a pressure transmitter calibration network, and to input the condition environment features of different modalities and the pressure data detected by the pressure transmitter in a preset time period into the pressure transmitter calibration network to perform prediction and calibration on the pressure data.

[0100] The application is described in detail above, and the principles and implementation modes of the application are described by applying specific examples. The above example is only used to help understand the method of the application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the application, the specific implementation mode and application range will be changed, and the above description should not be understood as a limitation on the application.

Claims

1. An adaptive calibration method for a pressure transmitter, characterized by, The method comprises the following steps: S1: collecting multiple sets of measurement sample data of the target pressure transmitter within a current time window; S2: constructing and training a measurement sample data encoder of a corresponding modal, encoding the measurement sample data as an environmental feature of the corresponding modal, and extracting common features of different modal environmental features; S3: using the common features as a condition, performing multi-scale reserved noise addition on the environmental features of the corresponding modal, training an error prediction network in the corresponding modal conditional diffusion model using the noise-added environmental features of the corresponding modal, and the error prediction network has the ability to fit the conditional environmental feature distribution; S4: performing multi-scale reserved noise addition on the environmental features of the corresponding modal, sampling using the error prediction network of the corresponding modal, and generating conditional environmental features of the corresponding modal; S5: constructing and training a pressure transmitter calibration network, inputting the conditional environmental features of different modal and the pressure data detected by the pressure transmitter within a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data. The method comprises the following steps: taking the common feature as a condition, performing multi-scale reserved noise on the environment feature of the corresponding modal, training an error prediction network in the corresponding modal conditional diffusion model using the environment feature of the corresponding modal after noise, the error prediction network has the ability to fit the conditional environment feature distribution, the conditional environment feature distribution is close to the common feature, comprising: taking the common feature As a condition of the variance prediction network in the diffusion model; the multi-scale reserved noise process is described as: ； in, for Environmental characteristics after adding noise; is the environmental feature without noise; is random noise, ; is the adjustment coefficient of the noise scale, which is used to control the noise scale; The variance prediction network of the corresponding modal diffusion model is trained using the environmental features with different modal and different noise scales, and the conditional environmental feature distribution of the corresponding modal is fitted, and the error prediction network is described as , is the training parameter of the network; The multi-scale reserved noise is added to the environment feature of the corresponding modality, and the error prediction network of the corresponding modality is used for sampling to generate the conditional environment feature of the corresponding modality, including: adding noise to the environment feature of the corresponding modality according to the multi-scale reserved noise process to obtain the noisy environment feature of the corresponding modality , and sampling to generate the conditional environment feature of the corresponding modality : ； ； where is the differential of is the first derivative of depends on the step size to control the noise level, and is the time step of the sampling; is the score function of is the differential of the time step is the interference term that satisfies the Itô process; and is the result obtained after sampling for a preset maximum time step .​​​​​ 2. An adaptive calibration method for a pressure transmitter as recited in claim 1, wherein, The method comprises the following steps:

3. An adaptive calibration method for a pressure transmitter as recited in claim 2, wherein, S1: collecting multiple sets of measurement sample data of the target pressure transmitter within a current time window; 4. An adaptive calibration method for a pressure transmitter as recited in claim 3, wherein, The method comprises constructing and training a measurement sample data encoder of the corresponding modality, encoding the measurement sample data as the environmental feature of the corresponding modality, and extracting the common features of the environmental features of different modalities, including taking the average value of the environmental features of different modalities as the initial common features. ,Will Considered as Update the weighted average weighting coefficient: ； wherein, is a weighting coefficient of the th modality environment feature; is an environment feature of the th modality; is an environment feature of the th modality; is a set containing all modality environment features; is a temperature parameter for controlling the smoothness of the weight; is a cosine similarity between two vectors; is a common feature at the current time; generating, by the weighting coefficient : ； ； In the formula, is a weighted mean value; is a weight adjustment coefficient; is a common feature at the next time point; Obtain the final common features through iteration .

5. The method for self-adapting calibration of a pressure transmitter according to claim 1, wherein, S2: constructing and training a measurement sample data encoder of a corresponding modal, encoding the measurement sample data as an environmental feature of the corresponding modal, and extracting common features of different modal environmental features; S3: using the common features as a condition, performing multi-scale reserved noise addition on the environmental features of the corresponding modal, training an error prediction network in the corresponding modal conditional diffusion model using the noise-added environmental features of the corresponding modal, and the error prediction network has the ability to fit the conditional environmental feature distribution; S4: performing multi-scale reserved noise addition on the environmental features of the corresponding modal, sampling using the error prediction network of the corresponding modal, and generating conditional environmental features of the corresponding modal; S5: constructing and training a pressure transmitter calibration network, inputting the conditional environmental features of different modal and the pressure data detected by the pressure transmitter within a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data.

6. An adaptive calibration method for a pressure transmitter as recited in claim 5, wherein, The mixed condition environment feature and the time sequence feature of the pressure data are input into the MLP, the pressure data at the current time is predicted, and a compensation strategy is selected, including: according to the deviation degree of the predicted pressure data relative to its historical sequence, the confidence of the current time predicted pressure data is evaluated, when the confidence is lower than the first preset threshold, the predicted pressure data and the original pressure measurement data are weighted and fused to output the final calibration value; when the confidence is lower than the second preset threshold, a low confidence alarm is issued, and the second preset threshold is less than the first preset threshold.

7. An adaptive calibration method for a pressure transmitter as recited in claim 5, wherein, The mixed condition environment feature and the time sequence feature of the pressure data are input into the MLP, the pressure data at the current time is predicted, and a compensation strategy is selected, including: when the predicted pressure data is greater than or equal to the first preset threshold, the current optimal compensation strategy is selected from the plurality of preset compensation strategies, or the parameters of the current compensation strategy are dynamically adjusted; the plurality of compensation strategies are optimized for different pressure ranges, environment change intensities or device running states.

8. An adaptive calibration system for a pressure transmitter for implementing an adaptive calibration method for a pressure transmitter according to any one of claims 1 to 7, characterized in that Comprise: A data acquisition module for collecting a plurality of measurement sample data of a target pressure transmitter in a current time window; A common feature extraction module for constructing and training a measurement sample data encoder corresponding to a modal, encoding the measurement sample data as an environment feature corresponding to the modal, and extracting common features of different modal environment features; A network training module for taking the common features as conditions, adding noise to the environment features of the corresponding modal in a multi-scale reservation manner, and training an error prediction network in a corresponding modal conditional diffusion model using the noise-added environment features of the corresponding modal, the error prediction network having the ability to fit the conditional environment feature distribution; A feature generation module for adding noise to the environment features of the corresponding modal in a multi-scale reservation manner, sampling using the error prediction network of the corresponding modal, and generating conditional environment features of the corresponding modal; A prediction calibration module for constructing and training a pressure transmitter calibration network, inputting the conditional environment features of different modal and the pressure data detected by the pressure transmitter in a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data. The mixed condition environment feature and the time sequence feature of the pressure data are input into the MLP, the pressure data at the current time is predicted, and a compensation strategy is selected, including: when the predicted pressure data is greater than or equal to the first preset threshold, the current optimal compensation strategy is selected from the plurality of preset compensation strategies, or the parameters of the current compensation strategy are dynamically adjusted; the plurality of compensation strategies are optimized for different pressure ranges, environment change intensities or device running states. Comprise: A data acquisition module for collecting a plurality of measurement sample data of a target pressure transmitter in a current time window; A common feature extraction module for constructing and training a measurement sample data encoder corresponding to a modal, encoding the measurement sample data as an environment feature corresponding to the modal, and extracting common features of different modal environment features; A network training module for taking the common features as conditions, adding noise to the environment features of the corresponding modal in a multi-scale reservation manner, and training an error prediction network in a corresponding modal conditional diffusion model using the noise-added environment features of the corresponding modal, the error prediction network having the ability to fit the conditional environment feature distribution; A feature generation module for adding noise to the environment features of the corresponding modal in a multi-scale reservation manner, sampling using the error prediction network of the corresponding modal, and generating conditional environment features of the corresponding modal; A prediction calibration module for constructing and training a pressure transmitter calibration network, inputting the conditional environment features of different modal and the pressure data detected by the pressure transmitter in a preset time period into the pressure transmitter calibration network to predict and calibrate the pressure data.

Images