Green ammonia process synthesis section abnormal condition early warning method

Abstract

The application provides a green ammonia process synthesis section abnormal condition early warning method. It comprises the following steps: collecting historical data, performing steady state determination on the working condition, and constructing a green ammonia process database under different working conditions; constructing a key variable library and a potential related variable library corresponding to the key variable, and performing feature extraction in the potential related variable library; constructing a BiLSTM-BiGRU prediction model; constructing a key variable threshold library under different working conditions; collecting real-time data and working condition conditions, and performing key variable abnormality early warning; and performing initial cause diagnosis and processing on the predicted abnormality. The application can realize targeted early warning of key variables in the synthesis section of the green ammonia process, and has universality under multiple steady state working conditions and complex load adjustment working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of green hydrogen ammonia synthesis, in particular to a green ammonia process synthesis section abnormal condition early warning method. BACKGROUND

[0002] Green ammonia, as a clean and efficient energy carrier, plays an important role in the modern chemical industry. Compared with the traditional Haber-Bosch ammonia synthesis process, green ammonia process has lower carbon emissions due to different sources of hydrogen. The research and application of green ammonia process has become an important development direction of the chemical industry. However, in the green ammonia production process, due to the volatility and intermittency of renewable energy such as wind and solar energy, it is difficult to match with the traditional continuous ammonia synthesis process. Therefore, in recent years, scholars have proposed a flexible ammonia synthesis technology applied to the green ammonia process. The flexible ammonia synthesis technology can support the regulation of production load, match with fluctuating wind and light resources, realize "production with source", and effectively improve the energy utilization efficiency and adaptability of the ammonia synthesis process. However, the ammonia synthesis reaction itself has a high risk, and the frequent load regulation in the green ammonia process brings many challenges to the safety of the production process: frequent load adjustment may cause changes in operating parameters such as temperature, pressure and flow, increase the instability of equipment operation, and cause operation out of control, equipment wear and tear, etc. The gas storage tank, compressor and other equipment may face additional stress and wear under frequent load adjustment, and may cause equipment failure or leakage and other safety problems. The above problems pose greater challenges to the safety of the green ammonia process. The green ammonia process needs to be synthesized in the reactor under high temperature and high pressure conditions, and ammonia is flammable and explosive and toxic to the human body. Once an explosion, leakage or other safety accidents occur, the consequences are unpredictable. Therefore, the requirement for abnormal condition management of the green ammonia process cannot be limited to real-time monitoring and diagnosis, but also needs to give early warning before the abnormal situation occurs, giving enough operation time to the on-site technical personnel, trying to avoid the occurrence of abnormalities and failures, and ensuring the safe and stable operation of the production process. SUMMARY

[0003] The present application provides a green ammonia process synthesis section abnormal condition early warning method to solve at least one of the above technical problems.

[0004] To solve the above problems, as one aspect of the present application, a green ammonia process synthesis section abnormal condition early warning method is provided, comprising:

[0005] Step 1: Collecting the historical data of the green ammonia process with time sequence, classifying according to different working conditions, and constructing the green ammonia process database under different working conditions;

[0006] Step 2: Construct a library of key variables in the synthesis section, and use the variables in the library as the output variables of the BiLSTM-BiGRU prediction model;

[0007] Step 3: For different output variables, construct a library of potentially relevant variables based on the process mechanism, and determine the model input variables by feature extraction of the relevant variable library through the multi-view aggregated adaptive encoder (MVAAE);

[0008] Step 4: Construct a BiLSTM-BiGRU hybrid neural network model for trend prediction of key parameters in the green ammonia process, and train it based on the different operating condition data sets obtained from the foregoing classification;

[0009] Step 5: Construct a library of early warning thresholds, and determine the early warning thresholds of the predicted variables under different operating conditions based on the foregoing different operating condition data sets;

[0010] Step 6: Collect real-time data of the green ammonia process and import them into the model. After preprocessing the data, use the models trained with different data sets to predict the key parameter values and change rates of the green ammonia process;

[0011] Step 7: According to the different output variables of the model, select the corresponding threshold value from the early warning threshold library constructed in step 5, and realize early warning of the process through an adaptive sliding window adjustment method based on a feedback mechanism to determine whether an abnormal situation occurs. If the system is determined to have no abnormal occurrence, return to step 6 to continuously warn the process; if an abnormal situation is determined to have occurred, an alarm is issued to alert the on-site operator of the upcoming abnormal situation, and the initial cause of the abnormality is diagnosed and treated.

[0012] Preferably, the green ammonia process database under different operating conditions in step 1 is classified as follows:

[0013] All data is classified into two categories: steady state and load switching state. The steady state category is further classified into the following 9 categories: production load 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%, and 110%.

[0014] The load switching category is divided into an ascending load adjustment group and a descending load adjustment group. The ascending load adjustment group is divided into the following 9 categories: production load 30%-40%, 40%-50%, 50%-60%, 60%-70%, 70%-80%, 80%-90%, 90%-100%, and 100%-110%.

[0015] The descending load adjustment group is divided into the following categories: production load 40%-30%, 50%-40%, 60%-50%, 70%-60%, 80%-70%, 90%-80%, 100%-90%, and 110%-100%.

[0016] Preferably, the green ammonia process synthesis section abnormal situation early warning method according to claim 1 is characterized by introducing a green ammonia stability index to determine whether the process is in a steady state, and for a single variable, the stability index is calculated as follows:

[0017]

[0018] Wherein, S A is the stability index of variable A, β A is the real-time change rate of variable A, ΔV A is the range of A within 10 minutes, V A,max -V A,min represents the range of variable A within the entire determination time range;

[0019] The stability index is calculated every 10 seconds, and when the S of all key variables is less than 0.1 within 10 minutes, the process is considered to be in a steady state.

[0020] Preferably, the synthesis section key variable library in step 2 specifically includes the following variables: synthesis column one bed temperature, synthesis column two bed temperature, synthesis column three bed temperature, synthesis column one bed pressure, synthesis column two bed pressure, synthesis column three bed pressure, and synthesis column outlet gas ammonia content.

[0021] Preferably, the potential correlation variable library in step 3 is specifically: synthesis column temperature potential correlation variable library, synthesis column pressure potential correlation variable library, and synthesis column outlet gas ammonia content potential correlation variable library; each library contains the following variables:

[0022] The synthesis column temperature potential correlation variable library contains 21 variables such as flash column overhead outlet flow, flash column overhead outlet temperature, and flash column bottom outlet pressure;

[0023] The synthesis column pressure potential correlation variable library contains 28 variables such as synthesis column inlet pressure, synthesis column outlet pressure, synthesis column inlet flow, and synthesis column outlet flow;

[0024] The synthesis column outlet gas ammonia content potential correlation variable library contains 22 variables such as synthesis column inlet ammonia content, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow, gas-liquid separation column overhead outlet temperature, gas-liquid separation column overhead outlet ammonia content, and gas-liquid separation column bottom outlet pressure.

[0025] Preferably, the multi-view angle aggregation adaptive encoder (MVAAE) in step 3 includes the following modules:

[0026] A multi-view angle encoder module for extracting features from multiple data sources of the ammonia synthesis section;

[0027] a graph neural network (GNN) fusion module for fusing the features extracted by the multi-view encoder modules to capture the inter-relationships between different data sources;

[0028] an attention mechanism module for dynamically adjusting the weights of different data sources in feature fusion according to their importance;

[0029] an adaptive dimension reduction module for reducing the dimension of the fused high-dimensional features while preserving key information.

[0030] Preferably, the multi-view encoder module includes multiple encoders, each for extracting features from different data sources; the data sources include: reaction temperature sensor data, reaction pressure sensor data, hydrogen-nitrogen ratio data, catalyst activity monitoring data, and reaction product concentration data; the encoder types include: multi-layer perceptron (MLP), convolutional neural network (CNN), and principal component analysis (PCA); wherein MLP is used to process reaction temperature sensor data and reaction pressure sensor data, CNN is used to process hydrogen-nitrogen ratio data and reaction product concentration data, and PCA is used to process catalyst activity monitoring data.

[0031] Preferably, the MVAAE model training module is optimized by minimizing the following loss function:

[0032] L = L recon + ωL reg

[0033] wherein L represents the total loss function, L recon represents the reconstruction loss function, L reg represents the regularization loss function, and ω is the regularization coefficient; wherein L recon is calculated as follows

[0034]

[0035] x represents the original input data, is the data reconstructed from the low-dimensional feature representation zlow by the decoder part of the adaptive dimension reduction module; L reg is calculated as follows

[0036]

[0037] α i represents the weight assigned to the i-th feature in the attention mechanism, and N is the number of features.

[0038] Preferably, in the pre-warning threshold library in step 5, each threshold has an index of variable name, load condition, threshold type and threshold value; wherein the load condition includes types consistent with the working condition types in the green ammonia process data set under the aforementioned different working conditions; and the threshold type includes fixed threshold value of variable value, adaptive dynamic threshold value of variable value, threshold value of variable value change rate and residual threshold value of difference between actual value and prediction of variable value; wherein the adaptive dynamic threshold value of variable value is obtained based on the Bayesian method, and the threshold value of variable value change rate and the residual threshold value are obtained based on the ROC method.

[0039] Preferably, when collecting real-time data of the green ammonia process in step 6, it is required to display the current load state on the central control display screen, and to display the current load in the steady state and the load switching task target being executed in the load switching.

[0040] Preferably, in step 7, the corresponding threshold value is selected from the threshold library to input the model for pre-warning; when the output variable is the temperature of the synthesis tower, the input threshold value is the adaptive dynamic threshold value of the variable value under the corresponding working condition, the threshold value of the variable value change rate and the residual threshold value; when the output variable is the pressure of the synthesis tower, the input threshold value is the fixed threshold value of the variable value under the corresponding working condition, the threshold value of the variable value change rate and the residual threshold value; when the output variable is the ammonia content of the gas out of the synthesis tower, the input threshold value is the fixed threshold value of the variable value under the corresponding working condition.

[0041] Preferably, when the process load adjustment task collected in step 6 cannot find a corresponding threshold value match in the green ammonia process database and the threshold library, a working condition interval adaptive determination method based on GNN is adopted to dynamically determine the working condition interval in which the current load is located by constructing a graph structure with process load and related process parameters as nodes and learning the correlation and similarity between the nodes in real time, so as to adjust the corresponding parameter prediction model and pre-warning strategy; for example, when the read real-time process load adjustment task in step 6 is the production load of 30%-70% and the process data and threshold value of the production load of 30%-70% working condition cannot be found in the green ammonia process database and the threshold library, the pre-warning task in step 7 can be divided into four plates of production load of 30%-40%, 40%-50%, 50%-60% and 60%-70% existing in the library, and the GNN is used to determine in which interval the production load adjustment task is located, and the pre-warning is performed in sections. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The whole operation process of the present application

[0043] Figure 2 The prediction and pre-warning result graph for the temperature pre-warning of the synthesis tower of the present application DETAILED DESCRIPTION

[0044] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0045] (1) Step 1: Collect historical data of the green ammonia process with time sequence, classify it according to different working conditions, and build a green ammonia process database under different working conditions;

[0046] The classification is as follows: all data are divided into two main categories: steady state and load switching state. The steady state category is further divided into the following 9 subcategories:

[0047] Production load: 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%, 110%;

[0048] Load switching is divided into load increase adjustment group and load decrease adjustment group. The load increase adjustment group is further divided into the following 9 categories: production load 30%-40%, 40%-50%, 50%-60%, 60%-70%, 70%-80%, 80%-90%, 90%-100%, and 100%-110%.

[0049] The load reduction adjustment groups are: production load 40%-30%, 50%-40%, 60%-50%, 70%-60%, 80%-70%, 90%-80%, 100%-90%, and 110%-100%.

[0050] In this classification, the green ammonia stability index is introduced to determine whether the process is in a steady state. For a single variable, the stability index is calculated as follows:

[0051]

[0052]

[0053] Among them, S A Let β be the stability exponent of variable A. A Let ΔV be the real-time rate of change of variable A. A Let V be the range of A over 10 minutes. A,max -V A,min This represents the range of variable A over the entire decision time period;

[0054] The stability index is calculated every 10 seconds. When the S-values ​​of all key variables are less than 0.1 within 10 minutes, the process is considered to be in a steady state.

[0055] (2) Step 2: Construct a key variable library for the synthesis section and use the variables in it as candidate output variables for the BiLSTM-BiGRU prediction model;

[0056] The key variable library in the synthesis section described in Step 2 specifically includes the following variables:

[0057] Synthesis column one bed temperature, synthesis column two bed temperature, synthesis column three bed temperature, synthesis column one bed pressure, synthesis column two bed pressure, synthesis column three bed pressure, synthesis column outlet gas ammonia content

[0058] (3) Step 3: For different output variables, a potential relevant variable library is constructed based on process mechanism, and model input variables are determined by feature extraction on the relevant variable library;

[0059] The potential relevant variable library described in Step 3 is specifically:

[0060] Synthesis column temperature potential relevant variable library, synthesis column pressure potential relevant variable library, synthesis column outlet gas ammonia content potential relevant variable library; Each library contains the following variables:

[0061] The synthesis column temperature potential relevant variable library contains the following 21 variables: synthesis column inlet temperature, synthesis column outlet temperature, synthesis column inlet flow rate, synthesis column outlet flow rate, hydrogen raw material flow rate, nitrogen raw material flow rate, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow rate, gas-liquid separation column overhead outlet temperature, gas-liquid separation column bottom outlet pressure, gas-liquid separation column bottom outlet temperature, gas-liquid separation column bottom outlet flow rate, flash tower overhead outlet pressure, flash tower overhead outlet flow rate, flash tower overhead outlet temperature, flash tower bottom outlet pressure, flash tower bottom outlet flow rate, flash tower bottom outlet flow rate, boiler steam temperature, boiler steam flow rate, feed four-stage compressor outlet temperature.

[0062] The synthesis column pressure potential relevant variable library contains the following 28 variables: synthesis column inlet pressure, synthesis column outlet pressure, synthesis column inlet flow rate, synthesis column outlet flow rate, feed one-stage compressor outlet temperature, feed one-stage compressor outlet pressure, feed two-stage compressor outlet temperature, feed two-stage compressor outlet pressure, feed three-stage compressor outlet temperature, feed three-stage compressor outlet pressure, feed three-stage compressor outlet temperature, feed three-stage compressor outlet pressure, feed four-stage compressor outlet temperature, feed four-stage compressor outlet pressure, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow rate, gas-liquid separation column overhead outlet temperature, gas-liquid separation column bottom outlet pressure, gas-liquid separation column bottom outlet temperature, gas-liquid separation column bottom outlet flow rate, flash tower overhead outlet pressure, flash tower overhead outlet flow rate, flash tower overhead outlet temperature, flash tower bottom outlet pressure, flash tower bottom outlet flow rate, flash tower bottom outlet flow rate, flash tower bottom outlet flow rate, boiler steam flow rate, cooling water temperature.

[0063] The potential correlation variable library of the ammonia content of the synthesis column outflow gas includes the following 22 variables: synthesis column inlet temperature, synthesis column inlet pressure, synthesis column inlet ammonia content, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow rate, gas-liquid separation column overhead outlet temperature, gas-liquid separation column overhead outlet ammonia content, gas-liquid separation column bottom outlet pressure, gas-liquid separation column bottom outlet temperature, gas-liquid separation column bottom outlet flow rate, gas-liquid separation column bottom outlet ammonia content, flash column overhead outlet pressure, flash column overhead outlet flow rate, flash column overhead outlet temperature, flash column overhead outlet ammonia content, flash column bottom outlet pressure, flash column bottom outlet flow rate, flash column bottom outlet flow rate, flash column bottom outlet ammonia content, ammonia product purity, hydrogen feed flow rate, and nitrogen feed flow rate.

[0064] The multi-view angle polymeric adaptive encoder (MVAAE) in step 3 includes the following modules:

[0065] A multi-view angle encoder module is used to extract features from multiple data sources of the ammonia synthesis section.

[0066] A graph neural network (GNN) fusion module is used to fuse the features extracted by the multi-view angle encoder module to capture the mutual relationship between different data sources.

[0067] An attention mechanism module is used to dynamically adjust the weight of different data sources in feature fusion according to their importance.

[0068] An adaptive dimension reduction module is used to reduce the dimension of the fused high-dimensional features while retaining key information.

[0069] The multi-view angle encoder module includes multiple encoders, each of which is used to extract features from different data sources; the data sources include reaction temperature sensor data, reaction pressure sensor data, hydrogen-nitrogen ratio data, catalyst activity monitoring data, and reaction product concentration data; the encoder types include multilayer perceptron (MLP), convolutional neural network (CNN), and principal component analysis (PCA); wherein MLP is used to process reaction temperature sensor data and reaction pressure sensor data, CNN is used to process hydrogen-nitrogen ratio data and reaction product concentration data, and PCA is used to process catalyst activity monitoring data.

[0070] The GNN module fuses the features of different data sources and captures their complex correlations; each feature z i is regarded as a node in the graph, and the edges between the nodes represent the mutual relationship between the data sources; the feature representation of each node is updated by the GNN to capture the global process information:

[0071]

[0072] The attention mechanism module dynamically adjusts the weight of each data source in the final feature fusion according to the importance of each data source, and calculates the attention score e of each feature i , and is normalized to weight a by a softmax function i :

[0073] e i =FFN(z i )

[0074]

[0075] The final aggregated feature is represented as:

[0076]

[0077] The adaptive dimension reduction module can reduce the fused high-dimensional features through a linear layer while retaining key information, facilitating subsequent analysis:

[0078] z low =g θ (z fusion )

[0079] In addition, the MVAAE model training module is optimized by minimizing the following loss function:

[0080] L=L recon +ωL reg

[0081] Where L represents the total loss function, L recon represents the reconstruction loss function, L reg represents the regularization loss function, and ω is the regularization coefficient; wherein the calculation formula of L recon is as follows

[0082]

[0083] x represents the original input data, is the data reconstructed from the low-dimensional feature representation zlow by the decoder part of the adaptive dimension reduction module; the calculation formula of L reg is as follows

[0084]

[0085] a i represents the weight assigned to the i-th feature in the attention mechanism, and N is the number of features.

[0086] (4) Step 4: Construct a BiLSTM-BiGRU hybrid neural network model for trend prediction of key parameters in the green ammonia process, and train it based on the different working condition data sets obtained from the aforementioned classification.

[0087] The BiLSTM-BiGRU hybrid model described in step 4, the model specific structure and implementation are: input the variable data after feature extraction into the model, the form of input data is

[0088] X = {x1, x2, …, x T}

[0089] x i is the input data vector at time step t; through the parallel BiLSTM and BiGRU layers, the time series data is predicted to obtain the output:

[0090]

[0091] The output result is input into a DENSE layer to generate weight parameters α t , which is obtained by updating through the back propagation algorithm with the minimum RMSE as the target, and the calculation formula of the weight function is

[0092] α t = sigmoid(W α ·x t +b α )

[0093] Where W α is the weight matrix, b α is the bias vector; after obtaining the weight parameter α t , the weight distribution result is input into the Multiply layer to weight the respective output results, and the calculation formula is as follows:

[0094] h Weighted,BiLSTM,t = α t ·h BiLSTM,t

[0095] h Weighted,BiGRU,t = (1-α t )·h BiGRU,t

[0096] Then to the Add layer to merge into the final output result in the BiLSTM-BiGRU parallel structure:

[0097] h Weighted,t = h Weighedt,BiGRU,t +h Weighedt,BiLSTM,t

[0098] H weighted,t = [h Weighted,1 , h Weighted , 2, …, h Weighted,T-1 , h Weighted,T ]

[0099] Finally, the final output value of the model is obtained through the full connection layer:

[0100] Y = [y1, y2,..., y T-1 , y T ]

[0101] where Y t ∈ R d represents the output result at time step t.

[0102] (5) Step 5: Constructing the early warning threshold library, based on the aforementioned different working condition data sets, determine the early warning threshold of the predicted variable under different working conditions;

[0103] In the early warning threshold library described in step 5, the index of each threshold has: variable name, load working condition, threshold type, threshold; Wherein, the type of load working condition is consistent with the type of working condition in the aforementioned green ammonia process data set under different working conditions; And the threshold type includes: fixed threshold of variable value, adaptive dynamic threshold of variable value, threshold of variable value change rate and residual threshold of difference between variable actual value and prediction; Wherein, the adaptive dynamic threshold of variable value is obtained based on the Bayesian method, the variable change rate threshold and the residual threshold are obtained based on the ROC method.

[0104] (6) Step 6: Collecting real-time data of green ammonia process and importing model, after data preprocessing, calling the model trained with different data sets to predict the green ammonia process key parameter value and change rate;

[0105] In step 6, the real-time data of green ammonia process is collected, which requires displaying the current load state in the central control display screen, displaying the current load in steady state, and displaying the load switching task target being executed when load switching.

[0106] (7) Step 7: According to the difference of the output variables of the model, select the corresponding threshold input from the early warning threshold library constructed in step 5, realize the early warning of the process through the adaptive sliding window adjustment method based on feedback mechanism, and determine whether there is abnormal situation. If the system is determined to have no abnormal situation, return to step 6 to continuously warn the process; If it is determined that an abnormal situation occurs, an alarm is issued to remind the on-site operator of the upcoming abnormal situation, and the initial cause leading to the abnormal situation is diagnosed and handled.

[0107] The corresponding threshold input model is selected from the threshold library for early warning in step 7, and the input threshold is: when the output variable is the temperature of the synthesis tower, the input threshold is the adaptive dynamic threshold of the variable value under the corresponding working condition, the variable value rate threshold and the residual error threshold; when the output variable is the pressure of the synthesis tower, the input threshold is the fixed threshold of the variable value under the corresponding working condition, the variable value rate threshold and the residual error threshold; when the output variable is the ammonia content of the gas out of the synthesis tower, the input threshold is the fixed threshold of the variable value under the corresponding working condition.

[0108] When the process load adjustment task collected in step 6 cannot find a corresponding threshold match in the green ammonia process database and the threshold library, a GNN-based working condition interval adaptive determination method is used to dynamically determine the working condition interval where the current load is located by constructing a graph structure with process load and related process parameters as nodes to learn the correlation and similarity between nodes in real time, thereby adjusting the corresponding parameter prediction model and early warning strategy; for example: the real-time process load adjustment task read in step 6 is the production load of 30%-70%, and the process data and threshold of the production load of 30%-70% working condition cannot be found in the green ammonia process database and the threshold library, the early warning task in step 7 can be divided into four plates of production load of 30%-40%, 40%-50%, 50%-60% and 60%-70% existing in the library, and the production load adjustment task is determined based on GNN to segment and perform early warning.

[0109] The adaptive sliding window adjustment method with feedback mechanism in step 7 of the method described in step 6 has the following specific embodiments:

[0110] ①Take the true positive rate (TPR) and false positive rate (FPR) of the early warning result as feedback indicators to dynamically adjust the size of the sliding window:

[0111]

[0112] Where FP is the number of false positives, FN is the number of false negatives, TP is the number of normal samples that are not alarmed, FN is the number of abnormal samples that are alarmed, FP+TN represents the total number of normal samples, and TP+FN represents the total number of fault samples;

[0113] ②Set the initial window size W0, the minimum window size W min , the maximum window size W max , the window

[0114] adjustment step P, window alarm step F and FPR, TPR index threshold T F , T T ;

[0115] ③Input the corresponding alarm threshold;

[0116] IV. If there are continuous F points in the window exceeding any threshold value described in right 6, it is determined that an abnormal situation occurs;

[0117] V. Combining the early warning results with the actual abnormal situation, calculate FPR and TPR;

[0118] VI. If the calculated FPR and TPR satisfy any one or more of TPR T , FPR F , it means that when the window size is too small, the window size needs to be increased, but cannot exceed the maximum window size:

[0119]

[0120] W t+1 = min(W t + P, W max )

[0121] If none of them is satisfied, it means that the current window size may be too large, and the window size can be reduced to improve the calculation efficiency, but

[0122] cannot exceed the maximum window size:

[0123] W t+1 = max(W t -P, W min )

[0124] The abnormality diagnosis and processing guidance described in step 7 is realized by combining SDG-HAZOP analysis, and the specific implementation is:

[0125] I. Perform HAZOP hazard and operability analysis on the green ammonia process, build an abnormal case library, and risk classify the abnormal situations in the case library according to the possible harm consequences, and develop disposal requirements for abnormal situations of different risk levels;

[0126] II. Abstract all measurable variables in the green ammonia process into nodes, and the relationship between two nodes into edges, on this basis, build the SDG signed directed graph G = (V, E) of the measurable variables in the green ammonia process, to reflect the relationship between variables, where V is the node set, and E is the directed edge set;

[0127] III. Assuming that the process variable of the early warning is temperature T, when it is determined that the system has an abnormality, find all nodes related to T through the signed directed graph, that is, all nodes satisfying (v i , v j ) and v j = T, and find all predecessor nodes of these variables through recursive method, and build a causal chain I;

[0128] ​(4) Starting from the abnormal variable T, each predecessor node is traced back in a recursive manner, until a variable without a predecessor node is found, which is the initial cause of the abnormality.

[0129] I: Put all nodes in I into a to-be-processed queue;

[0130] II: Take a variable from the queue, find all predecessor nodes, if the predecessor nodes also have abnormality, add them to the to-be-processed queue;

[0131] III: Traverse the entire queue, when the queue is empty, among all nodes in the queue, the variable without a predecessor node is the initial cause of the abnormality;

[0132] (5) After finding the initial cause, find the corresponding case and its risk classification result in the case library, and process the abnormality according to the disposal requirement corresponding to the risk classification result in (1).

[0133] The beneficial effects of the present application are:

[0134] (1) The present application is based on the stability index method, which determines the state of green ammonia historical data, classifies a large amount of process data clearly and accurately, and constructs a green ammonia historical database under different working conditions, providing a solid data support for green ammonia process early warning.

[0135] (2) The present application establishes a potential related variable library corresponding to different key variables, removes a large number of irrelevant variables based on process mechanism, and effectively improves the efficiency of feature variable extraction.

[0136] (3) The present application uses the multi-view aggregated adaptive encoder (MVAAE) method for feature engineering extraction from the potential related variable library, accurately extracts the strongly related variables corresponding to the key variables, removes the interference of redundant variables, and effectively improves the prediction effect of the prediction model.

[0137] (4) The present application constructs a diversified threshold library of variables under different working conditions and different threshold types, and selects different types of thresholds for different variables, which realizes the ability of each early warning task on the basis of ensuring the generalization ability of the model.

[0138] (5) The present application proposes a method for early warning task decomposition, which is used for early warning of load adjustment working conditions without matching working conditions in the case library, effectively improving the robustness and universality of the early warning method.

Claims

1. A method for green ammonia process synthesis section abnormal situation early warning, characterized in that: The method comprises the following steps: Step 1: Collecting historical data of the green ammonia process with time sequence, classifying according to different working conditions, and constructing a green ammonia process database under different working conditions; Step 2: Constructing a synthetic section key variable library, and taking the variables in the library as candidate output variables of the BiLSTM-BiGRU prediction model; Step 3: For different output variables, a potential related variable library is constructed based on the process mechanism, and a multi-view aggregation adaptive encoder is used to extract features of the related variable library to determine the input variables of the model; Step 4: Constructing a BiLSTM-BiGRU hybrid neural network model for predicting the trend of key parameters of the green ammonia process, and training the model based on the different working condition data sets obtained from the foregoing classification in the green ammonia process database; Step 5: Constructing a warning threshold library, and determining the warning threshold of the predicted variable under different working conditions based on the foregoing different working condition data sets; Step 6: Collecting real-time data of the green ammonia process and importing the model, and after preprocessing the data, calling the model trained using different data sets to predict the key parameter values and change rates of the green ammonia process; When the process load adjustment task collected in step 6 cannot find a corresponding threshold match in the green ammonia process database and the threshold library, an adaptive working condition interval determination method based on a graph neural network is used, a graph structure with process load and related process parameters as nodes is constructed, the correlation and similarity between nodes are learned in real time, and the working condition interval in which the current load is located is dynamically determined, so that the corresponding parameter prediction model and the warning strategy are adjusted; Step 7: According to the different output variables of the model, the corresponding threshold value input is selected from the warning threshold library constructed in step 5, the adaptive sliding window adjustment method based on the feedback mechanism is used to realize the warning of the process, and it is determined whether an abnormal situation occurs; if it is determined that no abnormal situation occurs, return to step 6 to continuously warn the process; If it is determined that an abnormal situation occurs, an alarm is sent to remind the on-site operator of the upcoming abnormal situation, and the initial cause of the abnormal situation is diagnosed and handled.

2. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: The green ammonia process database under different working conditions in step 1 is classified as follows: All data is classified into two categories: steady state and load switching state, and the steady state is classified into the following 9 categories: production load 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%, and 110%; The load switching class is divided into an ascending load adjustment group and a descending load adjustment group, the ascending load adjustment group is divided into the following 9 categories: production load 30%-40%, 40%-50%, 50%-60%, 60%-70%, 70%-80%, 80%-90%, 90%-100%, and 100%-110%; The descending load adjustment group is divided into: production load 40%-30%, 50%-40%, 60%-50%, 70%-60%, 80%-70%, 90%-80%, 100%-90%, and 110%-100%.

3. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein The stability index is introduced to determine whether the process is in a steady state. For a single variable, the stability index is calculated as follows: where S A is the stability index of variable A, β A is the real-time change rate of variable A, ΔV A is the range of A within 10 min, V A,max -V A,min represents the range of variable A within the entire judgment time range; The stability index is calculated every 10 seconds, and when the S of all key variables is less than 0.1 within 10 minutes, the process is considered to be in a steady state.

4. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: The key variable library in the synthesis section in step 2 specifically includes the following variables: synthesis column one bed temperature, synthesis column two bed temperature, synthesis column three bed temperature, synthesis column one bed pressure, synthesis column two bed pressure, synthesis column three bed pressure, and synthesis column outlet gas ammonia content.

5. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: The potential correlation variable library in step 3 is as follows: synthesis column temperature potential correlation variable library, synthesis column pressure potential correlation variable library, and synthesis column outlet gas ammonia content potential correlation variable library. Each library contains the following variables: The synthesis column temperature potential correlation variable library contains the following 21 variables: synthesis column inlet temperature, synthesis column outlet temperature, synthesis column inlet flow, synthesis column outlet flow, hydrogen raw material flow, nitrogen raw material flow, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow, gas-liquid separation column overhead outlet temperature, gas-liquid separation column bottom outlet pressure, gas-liquid separation column bottom outlet temperature, gas-liquid separation column bottom outlet flow, flash tower overhead outlet pressure, flash tower overhead outlet flow, flash tower overhead outlet temperature, flash tower bottom outlet pressure, flash tower bottom outlet flow, flash tower bottom outlet flow, boiler steam temperature, boiler steam flow, and feed four-stage compressor outlet temperature; The synthesis column pressure potential correlation variable library contains the following 28 variables: synthesis column inlet pressure, synthesis column outlet pressure, synthesis column inlet flow, synthesis column outlet flow, feed one-stage compressor outlet temperature, feed one-stage compressor outlet pressure, feed two-stage compressor outlet temperature, feed two-stage compressor outlet pressure, feed three-stage compressor outlet temperature, feed three-stage compressor outlet pressure, feed three-stage compressor outlet temperature, feed three-stage compressor outlet pressure, feed four-stage compressor outlet temperature, feed four-stage compressor outlet pressure, gas-liquid separation column overhead outlet pressure, gas-liquid separation column overhead outlet flow, gas-liquid separation column overhead outlet temperature, gas-liquid separation column bottom outlet pressure, gas-liquid separation column bottom outlet temperature, gas-liquid separation column bottom outlet flow, flash tower overhead outlet pressure, flash tower overhead outlet flow, flash tower overhead outlet temperature, flash tower bottom outlet pressure, flash tower bottom outlet flow, flash tower bottom outlet flow, boiler steam flow, and cooling water temperature; The potential correlation variable library of the ammonia content of the synthesis column outlet gas includes the synthesis column inlet temperature, the synthesis column inlet pressure, the synthesis column inlet ammonia content, the gas-liquid separation column overhead outlet pressure, the gas-liquid separation column overhead outlet flow, the gas-liquid separation column overhead outlet temperature, the gas-liquid separation column overhead outlet ammonia content, the gas-liquid separation column bottom outlet pressure, the gas-liquid separation column bottom outlet temperature, the gas-liquid separation column bottom outlet flow, the gas-liquid separation column bottom outlet ammonia content, the flash column overhead outlet pressure, the flash column overhead outlet flow, the flash column overhead outlet temperature, the flash column overhead outlet ammonia content, the flash column bottom outlet pressure, the flash column bottom outlet flow, the flash column bottom outlet flow, the flash column bottom outlet ammonia content, the ammonia product purity, the hydrogen feed flow, and the nitrogen feed flow, a total of 22 variables.

6. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: The multi-view polygenic adaptive encoder in step 3 includes the following modules: A multi-view encoder module for extracting features from multiple data sources of the ammonia synthesis section; A graph neural network fusion module for fusing the features extracted by the multi-view encoder module to capture the mutual relationship between different data sources; An attention mechanism module for dynamically adjusting the weight of different data sources in feature fusion according to their importance; An adaptive dimension reduction module for reducing the dimension of the fused high-dimensional features while retaining key information.

7. The green ammonia process synthesis section abnormal situation early warning method of claim 6, wherein: The multi-view encoder module includes multiple encoders, each for extracting features from different data sources; The data sources include reaction temperature sensor data, reaction pressure sensor data, hydrogen-nitrogen ratio data, catalyst activity monitoring data, and reaction product concentration data; the encoder types are multilayer perceptron, convolutional neural network, and principal component analysis; wherein the multilayer perceptron is used to process reaction temperature sensor data and reaction pressure sensor data, the convolutional neural network is used to process hydrogen-nitrogen ratio data and reaction product concentration data, and the principal component analysis is used to process catalyst activity monitoring data.

8. The green ammonia process synthesis section abnormal situation early warning method of claim 6, wherein: The multi-view polygenic adaptive encoder in step 3 optimizes the training model by minimizing the following loss function: L = L recon + ωL reg wherein L represents a total loss function, L recon represents a reconstruction loss function, L reg represents a regularization loss function, and ω is a regularization coefficient; wherein L recon The calculation formula of L is as follows x denotes the original input data, is the data reconstructed from the low-dimensional feature representation zlow by the decoder part of the adaptive dimensionality reduction module; L reg The computational formula is as follows a i denotes the weight assigned to the i-th feature in the attention mechanism, and N is the number of features.

9. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: In step 5, the index of each threshold in the pre-warning threshold library includes variable name, load condition, threshold type, and threshold; wherein the load condition includes the same types of load conditions as in the green ammonia process data set under different working conditions; and the threshold type includes fixed threshold of variable value, adaptive dynamic threshold of variable value, threshold of variable change rate, and residual threshold of the difference between actual value and prediction; wherein the adaptive dynamic threshold of variable value is obtained based on the Bayesian method, and the variable change rate threshold and the residual threshold are obtained based on the ROC method.

10. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: In step 6, when collecting real-time data of the green ammonia process, it is required to display the current load state on the central control display screen, and to display the current load in steady state and the load switching task target being executed during load switching.

11. The green ammonia process synthesis section abnormal situation early warning method of claim 1, wherein: The corresponding threshold input model is selected from the early warning threshold library constructed in step 5 for early warning in step 7, and the input threshold is: when the output variable is the temperature of the synthesis tower, the input threshold is the adaptive dynamic threshold of the variable value under the corresponding working condition, the variable value rate threshold and the residual error threshold; when the output variable is the pressure of the synthesis tower, the input threshold is the fixed threshold of the variable value under the corresponding working condition, the variable value rate threshold and the residual error threshold; when the output variable is the ammonia content of the gas out of the synthesis tower, the input threshold is the fixed threshold of the variable value under the corresponding working condition.

Images