Air preheater differential pressure prediction method and device, storage medium and computer device

Abstract

This invention discloses a method, apparatus, storage medium, and computer equipment for predicting differential pressure in air preheaters, relating to the field of air preheater safety technology. Its main purpose is to improve the accuracy and efficiency of differential pressure prediction in air preheaters. The method includes: responding to a differential pressure prediction signal from a target air preheater, determining the differential pressure prediction time step information for the target air preheater; determining a data acquisition time step based on the differential pressure prediction time step information; and acquiring, based on the data acquisition time step, the target air preheater's operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater; fusing the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data to obtain a fused time-series feature vector; and inputting the fused time-series feature vector into a preset differential pressure prediction model for differential pressure prediction to obtain a differential pressure time sequence corresponding to the differential pressure prediction time step.

Description

Technical Field

[0001] This invention relates to the field of air preheater safety technology, and in particular to a method, apparatus, storage medium, and computer equipment for predicting differential pressure in an air preheater. Background Technology

[0002] An air preheater, or air preheater for short, is a core heat exchange device in the boiler system of a thermal power generating unit. Its main function is to utilize the waste heat of the flue gas at the boiler tail end to heat the primary and secondary air required for combustion, thereby improving boiler combustion efficiency and reducing energy consumption. The operating status of the air preheater directly affects the safety, economy, and stability of the generating unit, and the air preheater pressure differential is a key indicator reflecting its operating status. It reflects the flow resistance encountered by the flue gas as it flows through the heat exchange elements of the air preheater. Therefore, to determine the operating status of the air preheater, it is first necessary to determine the air preheater pressure differential.

[0003] Currently, the air preheater pressure differential is typically determined by measuring the inlet and outlet flue gas pressures. However, this method, relying solely on flue gas pressure, results in low accuracy. Furthermore, manual measurement of flue gas pressure is time-consuming and labor-intensive, and is susceptible to errors in measuring tools, further reducing the accuracy of the air preheater pressure differential determination. Summary of the Invention

[0004] This invention provides a method, apparatus, storage medium, and computer equipment for predicting air preheater pressure difference, which mainly improves the accuracy and efficiency of air preheater pressure difference prediction.

[0005] According to a first aspect of the present invention, a method for predicting differential pressure in an air preheater is provided, comprising:

[0006] In response to the differential pressure prediction signal of the target air preheater, the differential pressure prediction time step information of the target air preheater is determined. Based on the differential pressure prediction time step information, the data acquisition time step is determined. Based on the data acquisition time step, the operating condition time sequence characteristic data, equipment status time sequence characteristic data, and heat exchange medium thermodynamic time sequence characteristic data of the target air preheater are acquired. The operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data are fused to obtain a fused time-series feature vector. The fused time-series feature vector is input into a preset differential pressure prediction model to predict differential pressure, thereby obtaining a differential pressure time series corresponding to the differential pressure prediction time step.

[0007] Optionally, the step of fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector includes: Based on the forward sequence information of the data acquisition time step, the forward operating condition time sequence association information of the operating condition time sequence feature data, the forward state time sequence association information of the equipment state time sequence feature data, and the forward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data are determined respectively. Based on the forward operating condition time series association information, the forward hidden operating condition vector of the operating condition time series feature data is determined; based on the forward state time series association information, the forward hidden state vector of the equipment state time series feature data is determined; based on the forward medium time series association information, the forward hidden medium vector of the heat exchange medium thermodynamic time series feature data is determined. Based on the reverse sequence information of the data acquisition time step, the backward operating condition time sequence association information of the operating condition time sequence feature data, the backward state time sequence association information of the equipment state time sequence feature data, and the backward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data are determined respectively. Based on the backward operating condition time series association information, the backward hidden operating condition vector of the operating condition time series feature data is determined; based on the backward state time series association information, the backward hidden state vector of the equipment state time series feature data is determined; based on the backward medium time series association information, the backward hidden medium vector of the heat exchange medium thermodynamic time series feature data is determined. The forward hidden condition vector, the forward hidden state vector, and the forward hidden medium vector are fused to obtain a forward fused vector. The backward hidden condition vector, the backward hidden state vector, and the backward hidden medium vector are fused to obtain a backward fused vector. The forward fused vector and the backward fused vector are concatenated to obtain the fused temporal feature vector.

[0008] Optionally, the step of fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector includes: Determine the operating condition time sequence feature vector corresponding to the operating condition time sequence feature data, the state time sequence feature vector corresponding to the equipment state time sequence feature data, and the medium time sequence feature vector corresponding to the heat exchange medium thermodynamic time sequence feature data, respectively. The amplitude of the operating condition vector, the amplitude of the state vector, and the amplitude of the medium vector corresponding to the medium timing feature vector are determined respectively. Based on the operating condition vector amplitude, the state vector amplitude, and the medium vector amplitude, the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient are determined respectively. Based on the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient, the operating condition time sequence feature weight, the state time sequence feature weight, and the medium time sequence feature weight are determined. Based on the operating condition weight reference coefficient, the operating condition time series feature vector is linearly transformed; based on the state weight reference coefficient, the state time series feature vector is linearly transformed; based on the medium weight reference coefficient, the medium time series feature vector is linearly transformed. Based on the operating condition time series feature weights, the state time series feature weights, and the medium time series feature weights, the linearly transformed operating condition time series feature vector, the linearly transformed state time series feature vector, and the linearly transformed medium time series feature vector are weighted and fused to obtain the fused time series feature vector.

[0009] Optionally, the preset differential pressure prediction model includes a feature extraction network and a differential pressure prediction network; The step of inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure, and obtaining a differential pressure time series corresponding to the differential pressure prediction time step, includes: Determine the historical pressure difference of the target air preheater in the previous time step corresponding to the current time step, and determine the historical pressure difference feature vector corresponding to the historical pressure difference. Input the fused time series feature vector into the feature extraction network for feature extraction. The historical pressure difference feature vector and the output vector of the feature extraction network are concatenated to obtain the concatenated feature, which is then input into the pressure difference prediction network for pressure difference time series prediction.

[0010] Optionally, after inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure and obtain a differential pressure time series corresponding to the differential pressure prediction time step, the method further includes: Obtain the normal differential pressure of the target air preheater under historical normal operating conditions, determine the differential pressure fluctuation coefficient of the target air preheater under normal operating conditions, and determine the standard differential pressure threshold of the target air preheater based on the normal differential pressure and the preset differential pressure fluctuation coefficient. Determine the slight abnormal fluctuation of the differential pressure of the target air preheater under slightly abnormal operating conditions, and determine the early warning benchmark threshold of the target air preheater based on the standard differential pressure threshold and the slight fluctuation of the differential pressure. Determine the amount of severe abnormal pressure fluctuation of the target air preheater under severe abnormal operating conditions, and determine the danger benchmark threshold of the target air preheater based on the standard pressure difference threshold and the amount of severe abnormal pressure fluctuation; Based on the pressure difference time series, the pressure difference development trend information is determined. Based on the pressure difference development trend information, the standard pressure difference threshold, the early warning benchmark threshold, the danger benchmark threshold, and the pressure difference at each time step in the pressure difference time series, the safety early warning mode of the target air preheater is determined, and a safety early warning is issued to the target air preheater based on the safety early warning mode.

[0011] Optionally, before fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain the fused time-series feature vector, the method further includes: Any one of the time-series feature data of the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data is taken as a target time-series feature data. The feature data at any moment in the target time-series feature data is taken as a target feature data to be detected. The range of feature data to be detected within a preset range corresponding to the target feature data to be detected is determined, and the average distance between the target feature data to be detected and the range of feature data to be detected is determined. Based on the mean distance, the neighborhood radius is dynamically determined, the data density within the neighborhood radius corresponding to the target feature data to be detected is determined, the density distribution information of the data density corresponding to each data in the target time-series feature data is determined, and a preset density threshold is dynamically determined based on the density distribution information. Determine whether the data density corresponding to the target feature data to be detected is less than the preset density threshold. If so, determine that the target feature data to be detected is abnormal data and remove the abnormal data from the target time series feature data. Identify the missing data in the target time-series feature data after removing abnormal data, and fill in the missing data in the target time-series feature data.

[0012] Optionally, before inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure and obtain a differential pressure time series corresponding to the differential pressure prediction time step, the method further includes: Construct a pre-defined initial differential pressure prediction model; Obtain a sample dataset, wherein the sample dataset includes time-series characteristic data of the operating conditions of the sample air preheater with differential pressure labels, time-series characteristic data of the equipment status, and time-series characteristic data of the heat exchange medium thermodynamics; The sample dataset is divided into a training set and a test set. The preset initial pressure difference prediction model is trained using the training set and tested using the test set. Finally, the preset initial pressure difference prediction model that meets the test conditions is taken as the preset pressure difference prediction model.

[0013] According to a second aspect of the present invention, an air preheater differential pressure prediction device is provided, comprising: The acquisition unit is used to respond to the differential pressure prediction signal of the target air preheater, determine the differential pressure prediction time step information of the target air preheater, determine the data acquisition time step based on the differential pressure prediction time step information, and acquire the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater based on the data acquisition time step. The fusion unit is used to fuse the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector. The prediction unit is used to input the fused time series feature vector into a preset differential pressure prediction model to perform differential pressure prediction and obtain a differential pressure time series corresponding to the differential pressure prediction time step.

[0014] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the above-described air preheater differential pressure prediction method.

[0015] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described air preheater differential pressure prediction method.

[0016] The present invention provides a method, apparatus, storage medium, and computer equipment for predicting air preheater differential pressure. Compared with the current method of measuring the inlet and outlet flue gas pressures of the air preheater and using the pressure difference between them as the air preheater differential pressure, the present invention predicts the air preheater differential pressure by comprehensively analyzing multiple factors such as the operating condition time-series characteristic data of the target air preheater, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data. This improves the prediction accuracy of air preheater differential pressure. By fusing the operating condition time-series characteristic data, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data, more implicit features in the data can be obtained, making fuller use of the data and increasing the accuracy of subsequent air preheater differential pressure prediction. The prediction of air preheater differential pressure is performed through a model, without any manual intervention, thereby improving the prediction efficiency and accuracy of air preheater differential pressure. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart of an air preheater differential pressure prediction method provided by an embodiment of the present invention is shown; Figure 2 A flowchart of another method for predicting air preheater differential pressure provided by an embodiment of the present invention is shown; Figure 3 This diagram illustrates the structure of an air preheater differential pressure prediction device according to an embodiment of the present invention. Figure 4 This invention provides a schematic diagram of another air preheater differential pressure prediction device according to an embodiment of the present invention. Figure 5 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0019] Currently, the method of measuring the inlet and outlet flue gas pressures of the air preheater and using the pressure difference between them as the air preheater pressure differential is based on a single factor, which leads to low accuracy in pressure differential prediction. In addition, manual measurement of flue gas pressure is time-consuming and labor-intensive.

[0020] To address the aforementioned problems, embodiments of the present invention provide a method for predicting the differential pressure of an air preheater, such as... Figure 1 As shown, the method includes: 101. In response to the differential pressure prediction signal of the target air preheater, determine the differential pressure prediction time step information of the target air preheater, determine the data acquisition time step based on the differential pressure prediction time step information, and acquire the operating condition time sequence characteristic data, equipment status time sequence characteristic data, and heat exchange medium thermodynamic time sequence characteristic data of the target air preheater based on the data acquisition time step.

[0021] Among them, the differential pressure prediction time step information refers to the future time step that needs to be predicted, such as the next 5 consecutive days; the data acquisition time step refers to the time step from a certain historical moment to the current moment, such as taking each day of the past 1 consecutive month as each data acquisition time step; the operating condition time series characteristic data includes but is not limited to the air preheater load; the equipment status time series characteristic data includes but is not limited to the air preheater current, the air preheater upper bearing temperature 1, the air preheater upper bearing temperature 2, the air preheater lower bearing temperature 1, and the air preheater lower bearing temperature 2; the heat exchange medium thermodynamic time series characteristic data includes but is not limited to the air preheater inlet flue gas temperature, the air preheater outlet flue gas temperature, the air preheater outlet temperature, the air preheater inlet primary air temperature, the air preheater outlet primary air temperature, the air preheater inlet secondary air pressure, the air preheater inlet secondary air temperature, the air preheater outlet secondary air pressure, and the air preheater outlet secondary air temperature.

[0022] In this embodiment of the invention, to accurately predict the air preheater pressure differential at multiple future time steps, it is first necessary to determine the data acquisition time step. Based on this, the method includes: determining the parameters and network structure depth of a preset pressure differential prediction model; determining the model complexity of the preset pressure differential prediction model based on the parameters and the network structure depth; acquiring the air preheater's operational characteristic data within a historical preset time period, wherein the operational characteristic data includes, but is not limited to, air preheater load, current, inlet flue gas temperature, outlet flue gas temperature, inlet velocity, outlet velocity, inlet pressure, and outlet pressure; determining the standard deviation, missing values, and outliers of the operational characteristic data; determining the data stability of the operational characteristic data based on the standard deviation, missing values, and outliers; determining the time step evaluation coefficient based on the model complexity and the data stability; and determining the data acquisition time step based on the time step evaluation coefficient and the pressure differential prediction time step information.

[0023] Specifically, the larger the number of model parameters or the deeper the network structure, the greater the model complexity. We can determine the weight coefficients corresponding to the number of model parameters and the depth of the network structure separately. Based on these weight coefficients, we sum the number of model parameters and the depth of the network structure to obtain a model complexity evaluation value. The larger this evaluation value, the greater the corresponding model complexity. Further, in determining data stability, we first determine the standard deviation of the operational feature data for the same type of data, such as inlet flue gas temperature. Based on the standard deviation, we determine the data volatility index parameter; for example, the larger the standard deviation, the larger the volatility index parameter. Then, we separately calculate the proportion of missing values ​​and outliers to the total data. Based on the proportion of missing values ​​and outliers, we determine the data anomaly index parameter. Finally, we perform a weighted summation of the data volatility index parameter and the anomaly index parameter, and determine the data stability based on the weighted summation result. Finally, we determine the weight coefficients corresponding to model complexity and data stability separately. Based on these weight coefficients, we perform a weighted summation of model complexity and data stability to obtain the time step evaluation coefficient. Furthermore, the data acquisition time step is determined according to the following formula. :

[0024] in, For differential pressure prediction time step, This is an additional buffer time step set according to actual needs. It should be noted that in addition to consecutive days, the time step can also be consecutive time points, etc.

[0025] Specifically, after determining the data acquisition time step, the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater are acquired based on the data acquisition time step. For example, if the data acquisition time step is the past 5 consecutive days, then it is necessary to acquire various time-series characteristic data of the air preheater from the current moment for the past 5 consecutive days. This embodiment of the invention reasonably determines the data acquisition time step by considering model complexity and data stability, which can avoid data computation redundancy caused by an excessively long data acquisition time step, and also avoid the problem of low pressure difference prediction accuracy due to insufficient information caused by an excessively short data acquisition time step. This embodiment of the invention comprehensively analyzes the data of various factors affecting pressure difference prediction, including operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data, which can ensure the comprehensiveness of the analysis and thus improve the prediction accuracy of air preheater pressure difference.

[0026] 102. The operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data are fused to obtain a fused time-series characteristic vector.

[0027] In this embodiment of the invention, for each type of time-series feature data—operating condition time-series feature data, equipment status time-series feature data, and heat exchange medium thermodynamic time-series feature data—a standard deviation is determined. Feature data whose difference between the time-series feature data and the standard deviation is greater than a preset threshold is considered outlier data and removed from the time-series feature data. Simultaneously, methods such as linear interpolation can be used to supplement missing data in the time-series feature data. The processed time-series feature data is then standardized.

[0028] Furthermore, to improve the prediction accuracy of differential pressure, it is first necessary to fuse the preprocessed operating condition time-series feature data, equipment status time-series feature data, and heat exchange medium thermodynamic time-series feature data. Based on this, step 102 specifically includes: determining the forward operating condition time-series association information of the operating condition time-series feature data, the forward state time-series association information of the equipment status time-series feature data, and the forward medium time-series association information of the heat exchange medium thermodynamic time-series feature data based on the forward operating condition time-series association information; determining the forward hidden operating condition vector of the operating condition time-series feature data based on the forward state time-series association information; determining the forward hidden state vector of the equipment status time-series feature data based on the forward medium time-series association information; and determining the forward hidden medium vector of the heat exchange medium thermodynamic time-series feature data based on the forward medium time-series association information; and determining the operating condition time-series feature data based on the reverse order information of the data acquisition time step. The system comprises: backward operating condition time-series correlation information of feature data; backward state time-series correlation information of equipment state time-series feature data; and backward medium time-series correlation information of heat exchange medium thermodynamic time-series feature data. Based on the backward operating condition time-series correlation information, a backward hidden operating condition vector of the operating condition time-series feature data is determined; based on the backward state time-series correlation information, a backward hidden state vector of the equipment state time-series feature data is determined; and based on the backward medium time-series correlation information, a backward hidden medium vector of the heat exchange medium thermodynamic time-series feature data is determined. The system then performs fusion processing on the forward hidden operating condition vector, the forward hidden state vector, and the forward hidden medium vector to obtain a forward fusion vector. Finally, the system performs concatenation processing on the forward fusion vector and the backward fusion vector to obtain the fused time-series feature vector.

[0029] Specifically, taking the determination of the forward hidden operating condition vector in the time-series feature data of operating conditions as an example, the feature data in the time-series feature data of operating conditions are sorted according to the time order from front to back in the data acquisition time step. After sorting, for the first preset number of feature data, the similarity between two adjacent feature data is calculated, and the mean similarity of all similarities is determined. Based on the mean similarity, the data association strength is determined. For example, the larger the mean similarity, the stronger the data association. This association strength is used as the forward operating condition time-series association information. Then, a feature extraction model, such as a CNN model or an encoder, is used to extract the forward hidden operating condition vector from the forward operating condition time-series association information.

[0030] Furthermore, taking the determination of the backward hidden operating condition vector in the time-series feature data of operating conditions as an example, the feature data in the time-series feature data of operating conditions are sorted according to the time sequence from back to front in the data acquisition time step. After sorting, for the first preset number of feature data, the similarity between two adjacent feature data is calculated, and the mean similarity of all similarities is determined. Based on the mean similarity, the data association strength is determined. For example, the larger the mean similarity, the stronger the data association. This association strength is used as the backward operating condition time-series association information. Then, a feature extraction model, such as a CNN model or an encoder, is used to extract the backward hidden operating condition vector from the backward operating condition time-series association information.

[0031] The process of determining the forward hidden state vector, forward hidden medium vector, backward hidden state vector, and backward hidden medium vector in this embodiment of the invention is the same as the process described above, and will not be repeated here.

[0032] Furthermore, the forward hidden operating condition vector, forward hidden state vector, and forward hidden medium vector are fused, such as by horizontal concatenation, to obtain a forward fused vector. Similarly, the backward hidden operating condition vector, backward hidden state vector, and backward hidden medium vector are fused, such as by horizontal concatenation, to obtain a backward fused vector. The forward and backward fused vectors are then concatenated to obtain a fused time-series feature vector. This embodiment of the invention, by fusing operating condition time-series feature data, equipment state time-series feature data, and heat exchange medium thermodynamic time-series feature data, can extract more latent features from the feature data, making fuller use of the data and resulting in more accurate differential pressure prediction results.

[0033] 103. Input the fused time series feature vector into the preset differential pressure prediction model to perform differential pressure prediction, and obtain the differential pressure time series corresponding to the differential pressure prediction time step.

[0034] In this embodiment of the invention, the preset differential pressure prediction model includes a feature extraction network and a differential pressure prediction network. The feature extraction network can be an encoder, and the differential pressure prediction network can be a decoder. Specifically, the method for differential pressure prediction using the preset differential pressure prediction model includes: determining the historical differential pressure of the target air preheater at the previous time step corresponding to the current time step, and determining the historical differential pressure feature vector corresponding to the historical differential pressure; inputting the fused time-series feature vector into the feature extraction network for feature extraction; concatenating the historical differential pressure feature vector with the output vector of the feature extraction network to obtain a concatenated feature; and inputting the concatenated feature into the differential pressure prediction network for differential pressure time-series prediction.

[0035] Specifically, the model adopts an Encoder-Decoder architecture with GRU (Gated Recurrent Unit) as its core. This architecture is suitable for sequence-to-sequence prediction tasks, where the Encoder processes the input sequence to extract time-dependent features, and the Decoder generates future prediction sequences based on the Encoder's feature sequences. GRU can effectively capture long sequence dependencies while maintaining high computational efficiency. The encoder is used to extract feature parameters from the past T time steps. The encoder architecture consists of the following: Input layer: input fused temporal feature vector; hidden layer dimension hidden_dim is set to 256; a Dropout layer is set to prevent overfitting. Encoder output: contains all temporal feature information from the past T time steps. Decoder (GRUDecoder): generates the pressure difference sequence for the next N time steps in an autoregressive manner based on the vector output by the encoder. Decoder initialization: the output vector of the last time step of the Encoder is set as the initial hidden state of the Decoder. The initial input x(1) of the first time step of the decoder is set to the true value of the pressure difference from the last time step of the past T time steps. The feature tensor C(i) output by the encoder at the current time step is concatenated with the pressure difference y(i-1) at the previous time step, and used as the input feature. This concatenation is then mapped to the input dimension of the decoder GRU through a Linear layer: gru_input(i). gru_input(i) and the encoder output feature s(i-1) from the previous time step are input into the unidirectional GRU to update the hidden state s(i) at the current time step. Pressure difference prediction output: The current hidden state s(i) is input into a fully connected layer to obtain the predicted pressure difference value y(i) at the i-th time step. y(i) is then used as one of the inputs for the next time step (i+1), and the above steps are repeated until prediction values ​​for N time steps are generated, ultimately forming a complete pressure difference prediction sequence. Predicting the air preheater pressure difference using this model requires no manual intervention, thus improving the prediction efficiency and accuracy of the air preheater pressure difference.

[0036] The air preheater pressure difference prediction method provided by this invention, compared with the current method of measuring the inlet and outlet flue gas pressures of the air preheater and using the pressure difference between them as the air preheater pressure difference, comprehensively analyzes multiple factors such as the operating condition time-series characteristic data of the target air preheater, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data to predict the air preheater pressure difference, which can improve the prediction accuracy of air preheater pressure difference. By fusing the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data, more implicit features in the data can be obtained, making fuller use of the data and thus increasing the accuracy of subsequent air preheater pressure difference prediction. The prediction of air preheater pressure difference is performed by a model, without the need for manual intervention, thereby improving the prediction efficiency and accuracy of air preheater pressure difference.

[0037] Furthermore, to better illustrate the above process of predicting the air preheater pressure difference, as a refinement and extension of the above embodiments, this invention provides another method for predicting air preheater pressure difference, such as... Figure 2 As shown, the method includes: 201. In response to the differential pressure prediction signal of the target air preheater, determine the differential pressure prediction time step information of the target air preheater, determine the data acquisition time step based on the differential pressure prediction time step information, and acquire the operating condition time sequence characteristic data, equipment status time sequence characteristic data, and heat exchange medium thermodynamic time sequence characteristic data of the target air preheater based on the data acquisition time step.

[0038] In this embodiment of the invention, after acquiring the time-series characteristic data of operating conditions, the time-series characteristic data of equipment status, and the time-series characteristic data of heat exchange medium, in order to improve data quality, before analyzing the above-mentioned characteristic data, it is necessary to perform anomaly detection and other processing on each of the time-series characteristic data. Based on this, the method includes: taking any one of the time-series characteristic data of operating conditions, the time-series characteristic data of equipment status, and the time-series characteristic data of heat exchange medium as a target time-series characteristic data; taking the characteristic data at any moment in the target time-series characteristic data as a target feature data to be detected; determining the range of feature data to be detected within a preset range corresponding to the target feature data to be detected; and determining the target feature data to be detected. The system detects the average distance between the target feature data and the target feature data within the specified range. Based on this average distance, it dynamically determines the neighborhood radius, determines the data density within the neighborhood radius corresponding to the target feature data, determines the density distribution information of the data density corresponding to each data point in the target time-series feature data, and dynamically determines a preset density threshold based on the density distribution information. It then determines whether the data density corresponding to the target feature data is less than the preset density threshold; if so, it identifies the target feature data as abnormal data and removes the abnormal data from the target time-series feature data. Finally, it identifies the missing data in the target time-series feature data after removing the abnormal data and completes the missing data in the target time-series feature data.

[0039] The preset range is set according to actual needs. Specifically, taking a certain feature data in a certain time series feature data as the target feature data to be detected as an example, if the data acquisition time step is 10 consecutive time points, then 5 data points before and after the target feature data to be detected can be taken as the range feature data to be detected. Calculate the absolute distance between the target feature data to be detected and each data point in the range feature data to be detected, and take the average of the distance sequence to obtain the distance mean. Multiply the distance mean by the dynamic adjustment coefficient to obtain the neighborhood radius, where the dynamic adjustment coefficient can be set based on the noise level of the data in the pre-processor. Determine a circular region centered on the target feature data to be detected with the neighborhood radius as the radius, and determine the data density of the feature data within this circular region. Thus, the data density of each feature data in the target time series feature data can be determined in the above manner, and the density mean and standard deviation of each data density can be determined. Based on the density mean and standard deviation, a preset density threshold is determined. If the data density of the target feature data to be detected in the target time series feature data is less than the preset density threshold, then the target feature data to be detected is determined as abnormal data, and this abnormal data is removed from the target time series feature data. If the data density of the target feature data to be detected is greater than or equal to a preset density threshold, the data is determined to be normal data. Since the air preheater is affected by factors such as geographical conditions and operating modes, the overall data distribution may dynamically change over time. In the process of abnormal data detection, this embodiment of the invention adapts to the distribution changes of the air preheater's operating data by dynamically determining the area radius and the preset density threshold, thereby improving the accuracy of abnormal data identification.

[0040] Furthermore, missing data in the target time-series feature data after removing outlier data is identified, and linear interpolation is used to fill in the missing data. This embodiment of the invention improves data quality by preprocessing the time-series feature data, thereby increasing the accuracy of subsequent differential pressure prediction and saving time spent analyzing outlier data.

[0041] 202. The operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data are fused to obtain a fused time-series characteristic vector.

[0042] In this embodiment of the invention, after performing anomaly handling and other processing on the operating condition time-series feature data, equipment status time-series feature data, and heat exchange medium thermodynamic time-series feature data, it is necessary to perform fusion processing on the preprocessed operating condition time-series feature data, equipment status time-series feature data, and heat exchange medium thermodynamic time-series feature data. Based on this, step 202 specifically includes: determining the operating condition time-series feature vector corresponding to the operating condition time-series feature data, the state time-series feature vector corresponding to the equipment status time-series feature data, and the medium time-series feature vector corresponding to the heat exchange medium thermodynamic time-series feature data; determining the operating condition vector amplitude of the operating condition time-series feature vector, the state vector amplitude of the state time-series feature vector, and the medium vector amplitude corresponding to the medium time-series feature vector; and based on the operating condition vector amplitude, the state vector amplitude, and the medium vector amplitude, respectively... Determine the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient; based on the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient, determine the operating condition time-series feature weight, the state time-series feature weight, and the medium time-series feature weight; based on the operating condition weight reference coefficient, perform a linear transformation on the operating condition time-series feature vector, based on the state weight reference coefficient, and based on the medium weight reference coefficient, perform a linear transformation on the medium time-series feature vector; based on the operating condition time-series feature weight, the state time-series feature weight, and the medium time-series feature weight, perform a weighted fusion of the linearly transformed operating condition time-series feature vector, the linearly transformed state time-series feature vector, and the linearly transformed medium time-series feature vector to obtain the fused time-series feature vector.

[0043] Specifically, firstly, a feature extraction model, such as a CNN model, is used to extract the operating condition time-series feature vectors corresponding to the operating condition time-series feature data, the state time-series feature vectors corresponding to the equipment state time-series feature data, and the medium time-series feature vectors corresponding to the heat exchange medium thermodynamic time-series feature data. During vector clustering, the square root of the sum of squares of each vector element in the operating condition time-series feature vector is first determined, and this square root is used as the amplitude of the operating condition vector. Simultaneously, the square root of the sum of squares of each vector element in the state time-series feature vector is determined, and this square root is used as the amplitude of the state vector. Similarly, the square root of the sum of squares of each vector element in the medium time-series feature vector is determined, and this square root is used as the amplitude of the medium vector. Then, based on the amplitudes of the operating condition vector, state vector, and medium vector, the sum of the squares of these amplitudes is determined, and the ratio of the square of the operating condition vector amplitude to the sum of the squares is used as the operating condition weight reference coefficient. Meanwhile, the ratio of the square of the state vector magnitude to the sum of the squares is used as the state weight reference coefficient. The sum of the operating condition weight reference coefficient and the state weight reference coefficient is determined, and 1 is subtracted from this sum to obtain the medium weight reference coefficient. .

[0044] Furthermore, based on the working condition weight reference coefficient State weight reference coefficient Medium weight reference coefficient The weights of the time-series features of the operating conditions are determined according to the following formula. State-time feature weights Medium timing characteristic weights :

[0045]

[0046]

[0047] Furthermore, the operating condition time series feature vector is processed according to the following formula. Perform a linear transformation:

[0048] At the same time, the state-time feature vector is processed according to the following formula. Perform a linear transformation:

[0049] Meanwhile, the medium timing feature vector is processed according to the following formula. Perform a linear transformation:

[0050] Furthermore, based on the weights of the operating condition time series features, the weights of the state time series features, and the weights of the medium time series features, the operating condition time series feature vector, the state time series feature vector, and the medium time series feature vector after linear transformation are weighted and summed to obtain the fused time series feature vector.

[0051] 203. Input the fused time series feature vector into the preset differential pressure prediction model to perform differential pressure prediction, and obtain the differential pressure time series corresponding to the differential pressure prediction time step.

[0052] In this embodiment of the invention, to improve the prediction accuracy of the preset differential pressure prediction model, it is first necessary to train and construct the preset differential pressure prediction model. Based on this, the method includes: constructing a preset initial differential pressure prediction model; obtaining a sample dataset, wherein the sample dataset includes time-series characteristic data of the operating conditions of a sample air preheater with differential pressure labels, time-series characteristic data of equipment status, and time-series characteristic data of the heat exchange medium thermodynamics; dividing the sample dataset into a training set and a test set, using the training set to train the preset initial differential pressure prediction model, and using the test set to test the trained preset initial differential pressure prediction model, and finally using the trained preset initial differential pressure prediction model that meets the test conditions as the preset differential pressure prediction model.

[0053] Specifically, during model training, a pre-defined initial differential pressure prediction model is first constructed, followed by the acquisition of a sample dataset. The dataset must contain all necessary files. The data is then converted to a format understandable by the pre-defined initial differential pressure prediction model. Finally, the model is trained and tested. Specifically, the dataset can be divided first: using randomness or a specific strategy (such as stratified sampling), the sample dataset is divided into training and testing sets. The model is then trained using the training set, and tested using the test set to evaluate its performance on unseen data. Precision, recall, and other metrics on the test set are calculated and recorded. If the model performance does not meet requirements, it can return to the training phase for further iterations or adjustments. This process yields a pre-defined differential pressure prediction model that meets the requirements. Simultaneously, a loss function can be constructed during model training, and the model is trained based on this loss function.

[0054] Finally, the fused time-series feature vector is input into the preset differential pressure prediction model for differential pressure prediction.

[0055] 204. Obtain the normal differential pressure of the target air preheater under historical normal operating conditions, determine the differential pressure fluctuation coefficient of the target air preheater under normal operating conditions, and determine the standard differential pressure threshold of the target air preheater based on the normal differential pressure and the preset differential pressure fluctuation coefficient.

[0056] 205. Determine the slight abnormal fluctuation of the differential pressure of the target air preheater under slightly abnormal operating conditions. Based on the standard differential pressure threshold and the slight fluctuation of differential pressure, determine the early warning benchmark threshold of the target air preheater.

[0057] 206. Determine the severe abnormal fluctuation of differential pressure in the target air preheater under severe abnormal operating conditions. Based on the standard differential pressure threshold and the severe abnormal fluctuation of differential pressure, determine the danger benchmark threshold of the target air preheater.

[0058] 207. Determine the pressure difference development trend information based on the pressure difference time series. Based on the pressure difference development trend information, standard pressure difference threshold, early warning benchmark threshold, danger benchmark threshold, and pressure difference at each time step in the pressure difference time series, determine the safety early warning mode for the target air preheater, and provide safety early warning for the target air preheater based on the safety early warning mode.

[0059] Specifically, the average normal differential pressure under historical normal operating conditions can be used as the normal differential pressure. The differential pressure fluctuation coefficient can be determined based on actual needs. The standard differential pressure threshold is Set the pressure differential for slight abnormal fluctuations based on actual needs. Then the warning baseline threshold for If the actual demand determines the amount of severe abnormal fluctuation in pressure difference to be... Then the danger threshold for Furthermore, if there are ≥1 time steps in the pressure difference time series where the value is in a certain range... The pressure differential is within a range, and the overall trend shows a slow upward trend, but it has not broken through... The safety warning method is to remind maintenance personnel to pay attention to the air preheater. If the differential pressure values ​​in the differential pressure time series are within ≥3 consecutive time steps... The range is defined, and the pressure differential shows a clear upward trend, with the final value breaking through... The safety warning method is to issue an alert to the operation and maintenance personnel. If the differential pressure values ​​in the predicted differential pressure sequence are within a certain range for ≥5 consecutive time steps... The interval, where the pressure difference values ​​at ≥1 time step are close to Furthermore, the pressure differential is trending upwards rapidly and is expected to break through [a certain threshold] in the short term. In this case, the safety warning method is to send an alarm to the operation and maintenance personnel. At the same time, the monitoring interface displays the abnormal time step, the predicted differential pressure value, and the trend curve of change, reminding the operation and maintenance personnel to check the air preheater's operating status in a timely manner.

[0060] According to another method for predicting air preheater pressure difference provided by the present invention, compared with the current method of measuring the inlet and outlet flue gas pressures of the air preheater and using the pressure difference between them as the air preheater pressure difference, the present invention predicts the air preheater pressure difference by comprehensively analyzing multiple factors such as the operating condition time-series characteristic data of the target air preheater, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data. This improves the prediction accuracy of air preheater pressure difference. By fusing the operating condition time-series characteristic data, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data, more implicit features in the data can be obtained, making fuller use of the data and thus increasing the accuracy of subsequent air preheater pressure difference prediction. The prediction of air preheater pressure difference is performed by a model, without any manual intervention, thereby improving the prediction efficiency and accuracy of air preheater pressure difference.

[0061] Furthermore, as Figure 1 In a specific implementation, this invention provides an air preheater differential pressure prediction device, such as... Figure 3 As shown, the device includes: an acquisition unit 31, a fusion unit 32, and a prediction unit 33.

[0062] The acquisition unit 31 can be used to respond to the differential pressure prediction signal of the target air preheater, determine the differential pressure prediction time step information of the target air preheater, determine the data acquisition time step based on the differential pressure prediction time step information, and acquire the operating condition time sequence characteristic data, equipment status time sequence characteristic data, and heat exchange medium thermodynamic time sequence characteristic data of the target air preheater based on the data acquisition time step.

[0063] The fusion unit 32 can be used to fuse the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector.

[0064] The prediction unit 33 can be used to input the fused time series feature vector into a preset differential pressure prediction model to perform differential pressure prediction, thereby obtaining a differential pressure time series corresponding to the differential pressure prediction time step.

[0065] In specific application scenarios, in order to fuse various time-series feature vectors, such as... Figure 4 As shown, the fusion unit 32 includes a determination module 321 and a fusion module 322.

[0066] The determining module 321 can be used to determine the forward operating condition time sequence association information of the operating condition time sequence feature data, the forward state time sequence association information of the equipment state time sequence feature data, and the forward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data based on the forward sequence information of the data acquisition time step.

[0067] The determining module 321 can also be used to determine the forward hidden operating condition vector of the operating condition time sequence feature data based on the forward operating condition time sequence association information, determine the forward hidden state vector of the equipment state time sequence feature data based on the forward state time sequence association information, and determine the forward hidden medium vector of the heat exchange medium thermodynamic time sequence feature data based on the forward medium time sequence association information.

[0068] The determining module 321 can also be used to determine the backward operating condition time sequence association information of the operating condition time sequence feature data, the backward state time sequence association information of the equipment state time sequence feature data, and the backward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data based on the reverse sequence information of the data acquisition time step.

[0069] The determining module 321 can also be used to determine the backward hidden operating condition vector of the operating condition time sequence feature data based on the backward operating condition time sequence association information, determine the backward hidden state vector of the equipment state time sequence feature data based on the backward state time sequence association information, and determine the backward hidden medium vector of the heat exchange medium thermodynamic time sequence feature data based on the backward medium time sequence association information.

[0070] The fusion module 322 can be used to fuse the forward hidden condition vector, the forward hidden state vector, and the forward hidden medium vector to obtain a forward fusion vector, and to fuse the backward hidden condition vector, the backward hidden state vector, and the backward hidden medium vector to obtain a backward fusion vector. The forward fusion vector and the backward fusion vector are then concatenated to obtain the fused temporal feature vector.

[0071] In specific application scenarios, in order to perform fusion processing on various time-series feature vectors, the fusion unit 32 also includes a linear transformation module 323 and a weighting module 324.

[0072] The determination module 321 can also be used to determine the operating condition time sequence feature vector corresponding to the operating condition time sequence feature data, the state time sequence feature vector corresponding to the equipment state time sequence feature data, and the medium time sequence feature vector corresponding to the heat exchange medium thermodynamic time sequence feature data.

[0073] The determining module 321 can also be used to determine the operating condition vector amplitude of the operating condition time sequence feature vector, the state vector amplitude of the state time sequence feature vector, and the medium vector amplitude corresponding to the medium time sequence feature vector.

[0074] The determining module 321 can also be used to determine the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient based on the operating condition vector amplitude, the state vector amplitude, and the medium vector amplitude, respectively.

[0075] The determining module 321 can also be used to determine the operating condition time sequence feature weight, the state time sequence feature weight, and the medium time sequence feature weight based on the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient.

[0076] The linear transformation module 323 can be used to perform linear transformation on the operating condition time-series feature vector based on the operating condition weight reference coefficient, to perform linear transformation on the state time-series feature vector based on the state weight reference coefficient, and to perform linear transformation on the medium time-series feature vector based on the medium weight reference coefficient.

[0077] The weighting module 324 can be used to perform weighted fusion of the linearly transformed operating condition time-series feature vector, the linearly transformed state time-series feature vector, and the linearly transformed medium time-series feature vector based on the operating condition time-series feature weight, the state time-series feature weight, and the medium time-series feature weight to obtain the fused time-series feature vector.

[0078] In specific application scenarios, the preset differential pressure prediction model includes a feature extraction network and a differential pressure prediction network; in order to predict the differential pressure time series, the prediction unit 33 includes a feature extraction module 331 and a differential pressure prediction module 332.

[0079] The feature extraction module 331 can be used to determine the historical pressure difference of the target air preheater in the previous time step corresponding to the current time step, and determine the historical pressure difference feature vector corresponding to the historical pressure difference, and input the fused time series feature vector into the feature extraction network for feature extraction.

[0080] The differential pressure prediction module 332 can be used to concatenate the historical differential pressure feature vector and the output vector of the feature extraction network to obtain concatenated features, and input the concatenated features into the differential pressure prediction network to perform differential pressure time series prediction.

[0081] In specific application scenarios, in order to provide safety warnings for air preheaters, the device also includes a warning unit 34.

[0082] The early warning unit 34 can be used to acquire the normal differential pressure of the target air preheater under historical normal operating conditions, determine the differential pressure fluctuation coefficient of the target air preheater under normal operating conditions, and determine the standard differential pressure threshold of the target air preheater based on the normal differential pressure and the preset differential pressure fluctuation coefficient; determine the slight abnormal fluctuation amount of the differential pressure of the target air preheater under slightly abnormal operating conditions, and determine the early warning benchmark threshold of the target air preheater based on the standard differential pressure threshold and the slight abnormal fluctuation amount; determine the severe abnormal fluctuation amount of the differential pressure of the target air preheater under severely abnormal operating conditions, and determine the danger benchmark threshold of the target air preheater based on the standard differential pressure threshold and the severe abnormal fluctuation amount; determine the differential pressure development trend information based on the differential pressure time series, and determine the safety early warning mode of the target air preheater based on the differential pressure development trend information, the standard differential pressure threshold, the early warning benchmark threshold, the danger benchmark threshold, and the differential pressure at each time step in the differential pressure time series, and provide a safety early warning to the target air preheater based on the safety early warning mode.

[0083] In specific application scenarios, in order to preprocess various time-series feature data, the device also includes a preprocessing unit 35.

[0084] The preprocessing unit 35 can be used to take any one of the time-series feature data of the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data as a target time-series feature data, take the feature data at any time in the target time-series feature data as a target feature data to be detected, determine the range of feature data to be detected within a preset range corresponding to the target feature data to be detected, and determine the average distance between the target feature data to be detected and the range of feature data to be detected; based on the average distance, dynamically determine the radius of the domain, determine the data density within the radius of the domain corresponding to the target feature data to be detected, determine the density distribution information of the data density corresponding to each data in the target time-series feature data, and dynamically determine a preset density threshold based on the density distribution information; determine whether the data density corresponding to the target feature data to be detected is less than the preset density threshold, and if so, determine that the target feature data to be detected is abnormal data and remove the abnormal data in the target time-series feature data; determine the missing data in the target time-series feature data after removing the abnormal data, and fill in the missing data in the target time-series feature data.

[0085] In specific application scenarios, in order to construct a preset differential pressure prediction model, the device also includes a model construction unit 36.

[0086] The model building unit 36 ​​can be used to build a preset initial differential pressure prediction model; obtain a sample dataset, wherein the sample dataset includes time-series characteristic data of the operating conditions of the sample air preheater with differential pressure labels, time-series characteristic data of the equipment status, and time-series characteristic data of the heat exchange medium thermodynamics; divide the sample dataset into a training set and a test set, use the training set to train the preset initial differential pressure prediction model, and use the test set to test the trained preset initial differential pressure prediction model, and finally use the trained preset initial differential pressure prediction model that meets the test conditions as the preset differential pressure prediction model.

[0087] It should be noted that other corresponding descriptions of the functional modules involved in the air preheater differential pressure prediction device provided in this embodiment of the invention can be found in the following references. Figure 1 The corresponding description of the method shown will not be repeated here.

[0088] Based on the above, Figure 1 Accordingly, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps: responding to a differential pressure prediction signal of a target air preheater, determining the differential pressure prediction time step information of the target air preheater; based on the differential pressure prediction time step information, determining a data acquisition time step; and based on the data acquisition time step, acquiring the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater; fusing the operating condition time-series characteristic data, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data to obtain a fused time-series characteristic vector; and inputting the fused time-series characteristic vector into a preset differential pressure prediction model for differential pressure prediction to obtain a differential pressure time sequence corresponding to the differential pressure prediction time step.

[0089] Based on the above, Figure 1 The method shown and as Figure 3 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 5As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: responding to a differential pressure prediction signal of the target air preheater, determining the differential pressure prediction time step information of the target air preheater; based on the differential pressure prediction time step information, determining a data acquisition time step; and based on the data acquisition time step, acquiring the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater; fusing the operating condition time-series characteristic data, the equipment status time-series characteristic data, and the heat exchange medium thermodynamic time-series characteristic data to obtain a fused time-series characteristic vector; and inputting the fused time-series characteristic vector into a preset differential pressure prediction model to perform differential pressure prediction, obtaining a differential pressure time sequence corresponding to the differential pressure prediction time step.

[0090] Through the technical solution of this invention, the invention predicts the air preheater pressure difference by comprehensively analyzing various factors such as the time-series characteristic data of the target air preheater's operating conditions, the time-series characteristic data of the equipment status, and the time-series characteristic data of the heat exchange medium's thermodynamics. This improves the prediction accuracy of the air preheater pressure difference. By fusing the time-series characteristic data of the operating conditions, the time-series characteristic data of the equipment status, and the time-series characteristic data of the heat exchange medium's thermodynamics, more implicit features in the data can be revealed, making fuller use of the data and thus increasing the accuracy of subsequent air preheater pressure difference prediction. The prediction of air preheater pressure difference is performed through a model without any human intervention, thereby improving the prediction efficiency and accuracy of air preheater pressure difference.

[0091] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting differential pressure in an air preheater, characterized in that, include: In response to the differential pressure prediction signal of the target air preheater, the differential pressure prediction time step information of the target air preheater is determined. Based on the differential pressure prediction time step information, the data acquisition time step is determined. Based on the data acquisition time step, the operating condition time sequence characteristic data, equipment status time sequence characteristic data, and heat exchange medium thermodynamic time sequence characteristic data of the target air preheater are acquired. The operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data are fused to obtain a fused time-series feature vector. The fused time-series feature vector is input into a preset differential pressure prediction model to predict differential pressure, thereby obtaining a differential pressure time series corresponding to the differential pressure prediction time step.

2. The method according to claim 1, characterized in that, The process of fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector includes: Based on the forward sequence information of the data acquisition time step, the forward operating condition time sequence association information of the operating condition time sequence feature data, the forward state time sequence association information of the equipment state time sequence feature data, and the forward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data are determined respectively. Based on the forward operating condition time series association information, the forward hidden operating condition vector of the operating condition time series feature data is determined; based on the forward state time series association information, the forward hidden state vector of the equipment state time series feature data is determined; based on the forward medium time series association information, the forward hidden medium vector of the heat exchange medium thermodynamic time series feature data is determined. Based on the reverse sequence information of the data acquisition time step, the backward operating condition time sequence association information of the operating condition time sequence feature data, the backward state time sequence association information of the equipment state time sequence feature data, and the backward medium time sequence association information of the heat exchange medium thermodynamic time sequence feature data are determined respectively. Based on the backward operating condition time series association information, the backward hidden operating condition vector of the operating condition time series feature data is determined; based on the backward state time series association information, the backward hidden state vector of the equipment state time series feature data is determined; based on the backward medium time series association information, the backward hidden medium vector of the heat exchange medium thermodynamic time series feature data is determined. The forward hidden condition vector, the forward hidden state vector, and the forward hidden medium vector are fused to obtain a forward fused vector. The backward hidden condition vector, the backward hidden state vector, and the backward hidden medium vector are fused to obtain a backward fused vector. The forward fused vector and the backward fused vector are concatenated to obtain the fused temporal feature vector.

3. The method according to claim 1, characterized in that, The process of fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector includes: Determine the operating condition time sequence feature vector corresponding to the operating condition time sequence feature data, the state time sequence feature vector corresponding to the equipment state time sequence feature data, and the medium time sequence feature vector corresponding to the heat exchange medium thermodynamic time sequence feature data, respectively. The amplitude of the operating condition vector, the amplitude of the state vector, and the amplitude of the medium vector corresponding to the medium timing feature vector are determined respectively. Based on the operating condition vector amplitude, the state vector amplitude, and the medium vector amplitude, the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient are determined respectively. Based on the operating condition weight reference coefficient, the state weight reference coefficient, and the medium weight reference coefficient, the operating condition time sequence feature weight, the state time sequence feature weight, and the medium time sequence feature weight are determined. Based on the operating condition weight reference coefficient, the operating condition time series feature vector is linearly transformed; based on the state weight reference coefficient, the state time series feature vector is linearly transformed; based on the medium weight reference coefficient, the medium time series feature vector is linearly transformed. Based on the operating condition time series feature weights, the state time series feature weights, and the medium time series feature weights, the linearly transformed operating condition time series feature vector, the linearly transformed state time series feature vector, and the linearly transformed medium time series feature vector are weighted and fused to obtain the fused time series feature vector.

4. The method according to claim 1, characterized in that, The preset differential pressure prediction model includes a feature extraction network and a differential pressure prediction network; The step of inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure, and obtaining a differential pressure time series corresponding to the differential pressure prediction time step, includes: Determine the historical pressure difference of the target air preheater in the previous time step corresponding to the current time step, and determine the historical pressure difference feature vector corresponding to the historical pressure difference. Input the fused time series feature vector into the feature extraction network for feature extraction. The historical pressure difference feature vector and the output vector of the feature extraction network are concatenated to obtain the concatenated feature, which is then input into the pressure difference prediction network for pressure difference time series prediction.

5. The method according to claim 1, characterized in that, After inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure and obtain a differential pressure time series corresponding to the differential pressure prediction time step, the method further includes: Obtain the normal differential pressure of the target air preheater under historical normal operating conditions, determine the differential pressure fluctuation coefficient of the target air preheater under normal operating conditions, and determine the standard differential pressure threshold of the target air preheater based on the normal differential pressure and the preset differential pressure fluctuation coefficient. Determine the slight abnormal fluctuation of the differential pressure of the target air preheater under slightly abnormal operating conditions, and determine the early warning benchmark threshold of the target air preheater based on the standard differential pressure threshold and the slight fluctuation of the differential pressure. Determine the amount of severe abnormal pressure fluctuation of the target air preheater under severe abnormal operating conditions, and determine the danger benchmark threshold of the target air preheater based on the standard pressure difference threshold and the amount of severe abnormal pressure fluctuation; Based on the pressure difference time series, the pressure difference development trend information is determined. Based on the pressure difference development trend information, the standard pressure difference threshold, the early warning benchmark threshold, the danger benchmark threshold, and the pressure difference at each time step in the pressure difference time series, the safety early warning mode of the target air preheater is determined, and a safety early warning is issued to the target air preheater based on the safety early warning mode.

6. The method according to claim 1, characterized in that, Before fusing the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain the fused time-series feature vector, the method further includes: Any one of the time-series feature data of the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data is taken as a target time-series feature data. The feature data at any moment in the target time-series feature data is taken as a target feature data to be detected. The range of feature data to be detected within a preset range corresponding to the target feature data to be detected is determined, and the average distance between the target feature data to be detected and the range of feature data to be detected is determined. Based on the mean distance, the neighborhood radius is dynamically determined, the data density within the neighborhood radius corresponding to the target feature data to be detected is determined, the density distribution information of the data density corresponding to each data in the target time-series feature data is determined, and a preset density threshold is dynamically determined based on the density distribution information. Determine whether the data density corresponding to the target feature data to be detected is less than the preset density threshold. If so, determine that the target feature data to be detected is abnormal data and remove the abnormal data from the target time series feature data. Identify the missing data in the target time-series feature data after removing abnormal data, and fill in the missing data in the target time-series feature data.

7. The method according to claim 1, characterized in that, Before inputting the fused time-series feature vector into a preset differential pressure prediction model to predict differential pressure and obtain the differential pressure time series corresponding to the differential pressure prediction time step, the method further includes: Construct a pre-defined initial differential pressure prediction model; Obtain a sample dataset, wherein the sample dataset includes time-series characteristic data of the operating conditions of the sample air preheater with differential pressure labels, time-series characteristic data of the equipment status, and time-series characteristic data of the heat exchange medium thermodynamics; The sample dataset is divided into a training set and a test set. The preset initial pressure difference prediction model is trained using the training set and tested using the test set. Finally, the preset initial pressure difference prediction model that meets the test conditions is taken as the preset pressure difference prediction model.

8. An air preheater differential pressure prediction device, characterized in that, include: The acquisition unit is used to respond to the differential pressure prediction signal of the target air preheater, determine the differential pressure prediction time step information of the target air preheater, determine the data acquisition time step based on the differential pressure prediction time step information, and acquire the operating condition time-series characteristic data, equipment status time-series characteristic data, and heat exchange medium thermodynamic time-series characteristic data of the target air preheater based on the data acquisition time step. The fusion unit is used to fuse the operating condition time-series feature data, the equipment status time-series feature data, and the heat exchange medium thermodynamic time-series feature data to obtain a fused time-series feature vector. The prediction unit is used to input the fused time series feature vector into a preset differential pressure prediction model to perform differential pressure prediction and obtain a differential pressure time series corresponding to the differential pressure prediction time step.

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

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images