Self-adaptive calibration method and system for pressure transmitter

Abstract

The invention relates to the technical field of pressure transmitters, in particular to a self-adaptive calibration method and system for a pressure transmitter. Comprising 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 to serve as environment characteristics of the corresponding mode, and extracting common characteristics of the environment characteristics of different modes; and S3, taking the common characteristics as conditions, performing multi-scale reserved noise addition on the environment characteristics of the corresponding modals, training an error prediction network in a corresponding modal condition diffusion model by using the environment characteristics subjected to noise addition of the corresponding modals, and enabling the error prediction network to have the capability of fitting condition environment characteristic distribution. According to the method, the generalization ability of the model to unseen or rare working conditions is improved through the latest diffusion model transformation thought and by adopting a multi-scale reserved noise adding and conditional sampling mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure transmitters, and in particular to an adaptive calibration method and system for a pressure transmitter. Background Art

[0002] As one of the key sensors in the fields of industrial automation, petrochemicals, aerospace, transportation, and environmental monitoring, pressure transmitters are responsible for converting measured pressure signals into standardized electrical signals and reporting them to the control system. In recent years, with the increase in process complexity and the harshness of the on-site environment, the problems of measurement deviation, zero drift, and sensitivity change in pressure transmitters during long-term operation have become increasingly prominent, directly affecting production safety and control accuracy. Due to the complex and changeable on-site environment of pressure transmitters, it is usually difficult to obtain high-quality labeled data covering all working conditions, which makes the calibration model based on supervised learning perform poorly when migrated to new scenarios. It also 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 of the Invention

[0003] In order to overcome the disadvantage of insufficient data for unseen or rare working conditions, the present invention provides an adaptive calibration method and system for a pressure transmitter.

[0004] The technical solution of the present invention is: an adaptive calibration method for a pressure transmitter, comprising the following steps: S1: Collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; S2: Construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; 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; S4: performing multi-scale preservation and noise addition on the environmental features of the corresponding modality, sampling using the error prediction network of the corresponding modality, and generating conditional environmental features of the corresponding modality; S5: Construct and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

[0005] Preferably, the collecting of multiple sets of measurement sample data of the target pressure transmitter within the current time window includes: obtaining multiple sets of measurement sample data of the target pressure transmitter within the current time window, the measurement sample data including instantaneous pressure data, ambient temperature data, temperature gradient data, installation posture data and historical compensation data.

[0006] Preferably, the measurement sample data encoder of the corresponding modality is constructed and trained, the measurement sample data is encoded as the environmental feature of the corresponding modality, and the common features of the environmental features of different modalities are extracted, including: constructing a measurement sample time series according to the time dimension based on the measurement sample data, the measurement sample time series includes an instantaneous pressure time series, an ambient temperature time series, a temperature gradient time series, an installation posture time series and a historical compensation time series, and using Transformer to extract time series features from the measurement sample time series, and the time series features are environmental features of different modalities.

[0007] Preferably, the construction and training of the measurement sample data encoder of the corresponding modality, encoding the measurement sample data as the environmental features of the corresponding modality, and extracting the common features of the environmental features of different modalities include: taking the average value of the environmental features of different modalities as the initial common features ,Will Considered as Update the weighted average weighting coefficients as follows: ; in, For the The weighting coefficients of the modal environment characteristics; For the Environmental characteristics of the modality; For the Environmental characteristics of the modality; is a set of all modal environment features; is the temperature parameter, which is used to control the smoothness of the weight; is the cosine similarity between two vectors; is the common characteristic of the current moment; The weighting coefficients are generated as follows : ; ; Where, is the weighted mean; is the weight adjustment coefficient; The common characteristics of the next moment; Obtain the final common features through iteration .

[0008] Preferably, the common features are used as conditions, the environmental features of the corresponding modalities are subjected to multi-scale retention and noise addition, and the environmental features after noise addition of the corresponding modalities are used to train the error prediction network in the conditional diffusion model of the corresponding modalities, wherein the error prediction network has the ability to fit the distribution of conditional environmental features, and the distribution of conditional environmental features is close to the common features, including: taking the common features as conditions, Considered as a condition of the variance prediction network in the diffusion model, the multi-scale preserving noise addition process is described as follows: ; 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 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 , are the training parameters of the network.

[0009] Preferably, the environmental features of the corresponding modality are subjected to multi-scale retention and noise addition, and the error prediction network of the corresponding modality is used for sampling to generate the conditional environmental features of the corresponding modality, including: the environmental features of the corresponding modality are subjected to noise addition according to the multi-scale retention and noise addition process to obtain the noisy environmental features of the corresponding modality. , and sample the corresponding modal conditional environment features as follows : ; ; Where, for The differential of for The first derivative of ; depends on the step size to control the noise level, and ; is the sampling time step; for The scoring function of is the time step The differential of To satisfy the interference term of the Ito process; after sampling the preset maximum time step, for .

[0010] Preferably, the pressure transmitter calibration network is constructed and trained, and the conditional environmental features of different modes and the pressure data detected by the pressure transmitter within a preset time period are sent to the pressure transmitter calibration network to predict and calibrate the pressure data, including: constructing and training the pressure transmitter calibration network, using self-Attention to aggregate the conditional environmental features of different modes, and calling the aggregated conditional environmental features mixed conditional environmental features, constructing a time series according to the time dimension for the pressure data detected by the pressure transmitter within a preset time period, using Transformer to extract the time series features of the pressure data, and sending the mixed conditional environmental features and the time series features of the pressure data into MLP to predict the pressure data at the current moment and select a compensation strategy.

[0011] Preferably, the mixed condition environment characteristics and the time series characteristics of the pressure data are fed into the MLP to predict the pressure data at the current moment and select a compensation strategy, including: evaluating the confidence of the predicted pressure data at the current moment based on the degree of deviation of the predicted pressure data relative to its historical sequence; when the confidence is lower than a first preset threshold, weightedly fusing the predicted pressure data with the original pressure measurement data to output a final calibration value; when the confidence is lower than a second preset threshold, issuing a low confidence alarm, and the second preset threshold is lower than the first preset threshold.

[0012] Preferably, the mixed condition environmental characteristics and the time series characteristics of the pressure data are fed into the MLP to predict the pressure data at the current moment and select a compensation strategy, including: when the predicted pressure data is greater than or equal to a first preset threshold, selecting the 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, severity of environmental changes or equipment operating conditions.

[0013] Preferably, an adaptive calibration system for a pressure transmitter comprises: A data acquisition module is used to collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; A common feature extraction module is used to construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; A network training module is configured to use the common features as conditions, perform multi-scale noise-preserving on the environmental features of the corresponding modality, and use the noise-preserving environmental features of the corresponding modality to train an error prediction network in the conditional diffusion model of the corresponding modality, wherein the error prediction network has the ability to fit the distribution of conditional environmental features; A feature generation module is used to perform multi-scale retention and noise addition on the environmental features of the corresponding modality, and use the error prediction network of the corresponding modality for sampling to generate conditional environmental features of the corresponding modality; The prediction and calibration module is used to build and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

[0014] The beneficial effects of the present invention are: 1. This invention uses a conditional diffusion model to model and sample environmental features in latent space. This allows it to generate conditional environmental features close to the true distribution for training or online completion when labeled samples are scarce or extreme operating conditions are not observed in the field. For pressure transmitters, this means that when encountering rare temperature gradients, special installation postures, or transient disturbances, the calibration network can be more informed with the generated conditional environmental features, thereby reducing the need for manual on-site calibration and frequent recalibration, saving maintenance costs and improving the confidence of continuous equipment operation. 2. This invention adopts a conservative noise addition method rather than a completely random generation method to preserve the large-scale physical information and temporal continuity in the original measurement during the sampling process, avoiding the generation of samples that violate physical laws and thus ensuring the reliability and consistency of the conditional environment characteristics used for training or online completion in a physical sense. 3. The present invention updates the common features through weighted iteration based on cosine similarity and sets an iterative convergence criterion, which can automatically reduce the impact of abnormal or noise modes during the common feature construction process, prevent abnormal samples from contaminating the global environment description, and thus ensure the adaptability and robustness of the calibration model in long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the adaptive calibration method for a pressure transmitter according to the present invention; Figure 2 Schematic diagram of the structure of the adaptive calibration system for pressure transmitter of the present invention. DETAILED DESCRIPTION

[0016] The present invention will now be described more fully hereinafter with reference to the accompanying drawings, in which presently preferred embodiments of the invention are shown. However, the invention can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided for thoroughness and completeness, and will fully convey the scope of the invention to those skilled in the art.

[0017] Example 1: An adaptive calibration method for a pressure transmitter, such as Figure 1 As shown, the following steps are included: S1: Collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; Acquire multiple groups of measurement sample data of the target pressure transmitter within the current time window, wherein the measurement sample data includes instantaneous pressure data, ambient temperature data, temperature gradient data, installation posture data and historical compensation data.

[0018] It should be explained that the measurement sample data includes the following: Instantaneous pressure data: the original sequence of instantaneous pressure values ​​recorded by the target pressure transmitter at each sampling moment; ambient temperature data: the sampling sequence of the temperature sensor at the installation location or the adjacent environment; temperature gradient data: the gradient sequence calculated by multi-point temperature sensors or by the temperature difference of adjacent sensors; installation attitude data: the time series of the installation angle, orientation or attitude measured by the three-axis acceleration / angular velocity sensor of the pressure transmitter; historical compensation data: the historical calibration / compensation coefficient of the pressure transmitter, the last calibration record or the compensation parameter sequence updated during operation.

[0019] S2: Construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; According to the measurement sample data, a measurement sample time series is constructed according to the time dimension. The measurement sample time series includes an instantaneous pressure time series, an ambient temperature time series, a temperature gradient time series, an installation posture time series and a historical compensation time series. Transformer is used to extract time series features from the measurement sample time series. The time series features are environmental features of different modes.

[0020] It should be explained that the original measurement sample data is uniformly preprocessed: Time alignment: interpolate or downsample the modal data according to a unified time base to ensure the formation of a synchronized measurement sample time series within the current time window; Filtering and denoising: Band-pass or low-pass filtering is used on the instantaneous pressure and temperature series to remove quantization noise and power frequency interference; Normalization: Each mode is normalized according to the historical mean and standard deviation or by sliding window, so that the numerical scales of different modes can be directly input into the subsequent encoder; Missing value processing: Linear or spline interpolation is used for short-term missing data. If the missing value exceeds the threshold, the time window is marked as defective data and a supplementary sampling or fallback strategy is triggered; The obtained synchronized modal data are constructed into a measurement sample time series according to the time dimension, and the instantaneous pressure time series, ambient temperature time series, temperature gradient time series, installation posture time series and historical compensation time series are obtained. The fixed-dimensional time series feature vector is generated by Transformer through pooling in the time dimension (for example, taking the last time step output or using self-attention weighted pooling) or global average pooling. This vector serves as the environmental feature of the corresponding modality.

[0021] Take the average value of different modal environment features as the initial common feature ,Will Considered as Update the weighted average weighting coefficients as follows: ; in, For the The weighting coefficients of the modal environment characteristics; For the Environmental characteristics of the modality; For the Environmental characteristics of the modality; is a set of all modal environment features; is the temperature parameter, which is used to control the smoothness of the weight; is the cosine similarity between two vectors; is the common characteristic of the current moment; The weighting coefficients are generated as follows : ; ; Where, is the weighted mean; is the weight adjustment coefficient; The common characteristics of the next moment; Obtain the final common features through iteration .

[0022] 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, Calculate the arithmetic mean of all modal environment features as the initial common feature: ; in is a set containing all modes, is the modal number, Assigned to , as the initial value of iteration; 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; 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; 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.

[0023] 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; 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. 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.

[0024] 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; 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: ; 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 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 , are the training parameters of the network.

[0025] It needs to be explained that when near hour, Closer (The proportion of original content retained is high); when Near 0 o'clock, Closer to pure Gaussian noise (strong perturbation); to enhance multi-scale properties, each sample in the training set is randomly sampled (evenly or according to a preset distribution) to obtain a set of noise samples at multiple scales; Error Prediction Network The goal is to extract the noise from the sample Common conditions With time step Recover and predict the noise component , so the following mean square error loss is used: ; During training, train each modality separately (i.e., using an independent error prediction network for each mode), using multi-layer MLP as The network input is the concatenated vector; the conditional access method uses direct concatenation and feeds 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.

[0026] S4: performing multi-scale preservation and noise addition on the environmental features of the corresponding modality, sampling using the error prediction network of the corresponding modality, and generating conditional environmental features of the corresponding modality; The environmental features of the corresponding modality are denoised according to the multi-scale preservation denoising process to obtain the noisy environmental features of the corresponding modality. , and sample the corresponding modal conditional environment features as follows : ; ; Where, for The differential of for The first derivative of ; depends on the step size to control the noise level, and ; is the sampling time step; for The scoring function of is the time step The differential of To satisfy the interference term of the Ito process; after sampling the preset maximum time step, for .

[0027] It should be explained that the conditional environmental features are generated by sampling through the above process, and the conditional environmental features are transformed through the common features so that the distribution of the conditional environmental features is close to the common features, thereby improving the interference caused by the missing of single modal data or incorrect measurement on the subsequent network prediction value.

[0028] S5: Construct and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

[0029] A pressure transmitter calibration network was constructed and trained. Self-Attention was used to aggregate the conditional environmental features of different modalities. The aggregated conditional environmental features were called mixed conditional environmental features. The pressure data detected by the pressure transmitter within a preset time period was constructed into a time series according to the time dimension. Transformer was used to extract the temporal features of the pressure data. The mixed conditional environmental features and the temporal features of the pressure data were fed into the MLP to predict the pressure data at the current moment and select a compensation strategy.

[0030] The confidence level of the predicted pressure data at the current moment is evaluated based on the degree of deviation of the predicted pressure data from its historical sequence. When the confidence level is lower than a first preset threshold, the predicted pressure data and the original pressure measurement data are weightedly fused to output a final calibration value. When the confidence level is lower than a second preset threshold, a low confidence alarm is issued, and the second preset threshold is lower than the first preset threshold.

[0031] It should be explained that the confidence of the predicted pressure data output after MLP based on the temporal characteristics of the pressure data obtained by Transformer based on the mixed conditional environmental characteristics and the pressure measurement time series is quantified and weighted fused by three types of uncertainty: the first is the deviation of the predicted value from the historical series (mapped to the confidence by z-score), the second is the model prediction uncertainty (estimated by the output variance or model integration / MC-Dropout), and the third is the conditional environment consistency (a measure of the consistency 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 fused using 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 deviation threshold), the current optimal compensation strategy is selected from a variety of pre-set compensation strategies based on the historical benefit / cost / risk score, or the compensation strategy parameters are dynamically adjusted according to the confidence, taking into account calibration accuracy, execution cost and equipment safety.

[0032] When the predicted pressure data is greater than or equal to a first preset threshold, the current optimal compensation strategy is selected from a 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, severity of environmental changes or equipment operating conditions.

[0033] Example 2: Based on Example 1, an adaptive calibration system for a pressure transmitter, such as Figure 2 Shown, including: A data acquisition module is used to collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; A common feature extraction module is used to construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; A network training module is configured to use the common features as conditions, perform multi-scale noise-preserving on the environmental features of the corresponding modality, and use the noise-preserving environmental features of the corresponding modality to train an error prediction network in the conditional diffusion model of the corresponding modality, wherein the error prediction network has the ability to fit the distribution of conditional environmental features; A feature generation module is used to perform multi-scale retention and noise addition on the environmental features of the corresponding modality, and use the error prediction network of the corresponding modality for sampling to generate conditional environmental features of the corresponding modality; The prediction and calibration module is used to build and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

[0034] The above is a detailed introduction to the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, based on the idea of ​​the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An adaptive calibration method for a pressure transmitter, characterized in that: The following steps are involved: S1: Collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; S2: Construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; 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; S4: performing multi-scale preservation and noise addition on the environmental features of the corresponding modality, sampling using the error prediction network of the corresponding modality, and generating conditional environmental features of the corresponding modality; S5: Construct and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

2. The adaptive calibration method for a pressure transmitter according to claim 1, characterized in that: The collecting of multiple sets of measurement sample data of the target pressure transmitter within the current time window includes: obtaining multiple sets of measurement sample data of the target pressure transmitter within the current time window, the measurement sample data including instantaneous pressure data, ambient temperature data, temperature gradient data, installation posture data and historical compensation data.

3. The adaptive calibration method for a pressure transmitter according to claim 2, characterized in that: The method comprises constructing and training a measurement sample data encoder of the corresponding modality, encoding the measurement sample data as the environmental features of the corresponding modality, and extracting common features of environmental features of different modalities, including: constructing a measurement sample time series according to the time dimension based on the measurement sample data, wherein the measurement sample time series includes an instantaneous pressure time series, an ambient temperature time series, a temperature gradient time series, an installation posture time series, and a historical compensation time series, and using a Transformer to extract time series features from the measurement sample time series, wherein the time series features are environmental features of different modalities.

4. The adaptive calibration method for a pressure transmitter according to claim 3, characterized in that: 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 coefficients as follows: ； in, For the The weighting coefficients of the modal environment characteristics; For the Environmental characteristics of the modality; For the Environmental characteristics of the modality; is a set of all modal environment features; is the temperature parameter, which is used to control the smoothness of the weight; is the cosine similarity between two vectors; is the common characteristic of the current moment; The weighting coefficients are generated as follows : ； ； Where, is the weighted mean; is the weight adjustment coefficient; The common characteristics of the next moment; Obtain the final common features through iteration .

5. The adaptive calibration method for a pressure transmitter according to claim 1, characterized in that: The method uses the common features as conditions, performs multi-scale retention and noise addition on the environmental features of the corresponding modalities, and uses the environmental features after noise addition on the corresponding modalities to train the error prediction network in the conditional diffusion model of the corresponding modalities, wherein the error prediction network has the ability to fit the distribution of conditional environmental features, and the distribution of conditional environmental features is close to the common features, including: Considered as a condition of the variance prediction network in the diffusion model, the multi-scale preserving noise addition process is described as follows: ； in, for Environmental characteristics after adding noise; is the environmental characteristic 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 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 , are the training parameters of the network.

6. The adaptive calibration method for a pressure transmitter according to claim 1, characterized in that: The environmental features of the corresponding modality are subjected to multi-scale retention and noise addition, and the error prediction network of the corresponding modality is used for sampling to generate the conditional environmental features of the corresponding modality, including: the environmental features of the corresponding modality are subjected to noise addition according to the multi-scale retention and noise addition process to obtain the noisy environmental features of the corresponding modality. , and sample the corresponding modal conditional environment features as follows : ； ； Where, for The differential of for The first derivative of ; depends on the step size to control the noise level, and ; is the sampling time step; for The scoring function of is the time step The differential of To satisfy the interference term of the Ito process; after sampling the preset maximum time step, for .

7. The adaptive calibration method for a pressure transmitter according to claim 1, characterized in that: The pressure transmitter calibration network is constructed and trained, and the conditional environmental features of different modes and the pressure data detected by the pressure transmitter within a preset time period are sent to the pressure transmitter calibration network to predict and calibrate the pressure data, including: constructing and training the pressure transmitter calibration network, using self-Attention to aggregate the conditional environmental features of different modes, and calling the aggregated conditional environmental features mixed conditional environmental features, constructing a time series according to the time dimension of the pressure data detected by the pressure transmitter within the preset time period, using Transformer to extract the time series features of the pressure data, and sending the mixed conditional environmental features and the time series features of the pressure data into MLP to predict the pressure data at the current moment and select a compensation strategy.

8. The adaptive calibration method for a pressure transmitter according to claim 7, characterized in that: The mixed condition environment characteristics and the time series characteristics of the pressure data are fed into the MLP to predict the pressure data at the current moment and select a compensation strategy, including: evaluating the confidence of the predicted pressure data at the current moment based on the degree of deviation of the predicted pressure data relative to its historical sequence; when the confidence is lower than a first preset threshold, weightedly fusing the predicted pressure data with the original pressure measurement data to output a final calibration value; when the confidence is lower than a second preset threshold, issuing a low confidence alarm, and the second preset threshold is lower than the first preset threshold.

9. The adaptive calibration method for a pressure transmitter according to claim 7, characterized in that: The mixed condition environmental characteristics and the time series characteristics of the pressure data are fed into the MLP to predict the pressure data at the current moment and select a compensation strategy, including: when the predicted pressure data is greater than or equal to a first preset threshold, selecting the 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, severity of environmental changes, or equipment operating conditions.

10. An adaptive calibration system for a pressure transmitter, used to implement the adaptive calibration method for a pressure transmitter according to any one of claims 1 to 9, characterized in that: include: A data acquisition module is used to collect multiple groups of measurement sample data of the target pressure transmitter within the current time window; A common feature extraction module is used to construct and train a measurement sample data encoder for the corresponding modality, encode the measurement sample data as the environmental features of the corresponding modality, and extract common features of environmental features of different modalities; A network training module is configured to use the common features as conditions, perform multi-scale noise-preserving on the environmental features of the corresponding modality, and use the noise-preserving environmental features of the corresponding modality to train an error prediction network in the conditional diffusion model of the corresponding modality, wherein the error prediction network has the ability to fit the distribution of conditional environmental features; A feature generation module is used to perform multi-scale retention and noise addition on the environmental features of the corresponding modality, and use the error prediction network of the corresponding modality for sampling to generate conditional environmental features of the corresponding modality; The prediction and calibration module is used to build and train a pressure transmitter calibration network, and send the conditional environmental characteristics 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.

Images