Ammonia refrigeration system risk assessment method

Abstract

The invention discloses an ammonia refrigeration system risk assessment method, which is characterized in that a plurality of data sources are integrated, a more comprehensive and accurate safety assessment model is established, and potential risks are found in time and early warning is given out. The method comprises the following steps: 1, preprocessing traditional sensor data, category data, gas concentration data and video monitoring data; 2, dimension reduction is conducted on the preprocessed data, then the data are combined into a data set, and the data set is divided into a training set and a test set; 3, performing random forest model training and isolated forest model training according to the data set, and generating an online model in combination with a random forest model and an isolated forest model; and 4, performing risk assessment by using the online model ammonia refrigeration system. The method belongs to the technical field of refrigeration system protection.

Description

Technical Field

[0001] The present invention belongs to the technical field of refrigeration system protection, and more specifically, relates to a risk assessment method for an ammonia refrigeration system. Background Art

[0002] With the wide application of ammonia refrigeration systems in multiple industries such as food processing, refrigeration, and air conditioning, the importance of their safe operation has become increasingly prominent; ammonia, as a refrigerant, has significant advantages in terms of efficiency and environmental protection, capable of providing efficient heat exchange effects, and its low global warming potential (GWP) makes it a more environmentally friendly choice. However, ammonia itself also has certain safety hazards, especially its toxicity and flammability. Once a leakage or system failure occurs, it may pose a great threat to operators, the surrounding environment, and equipment. Therefore, how to effectively ensure the safe operation of ammonia refrigeration systems has become an urgent problem to be solved in the current refrigeration field. Summary of the Invention

[0003] The main purpose of the present invention is to provide a risk assessment method for an ammonia refrigeration system, which integrates multiple data sources to establish a more comprehensive and accurate safety assessment model, and timely discovers potential risks and issues warnings.

[0004] According to the first aspect of the present invention, a risk assessment method for an ammonia refrigeration system is provided, including the following steps:

[0005] Step 1: Preprocess traditional sensor data, category data, gas concentration data, and video surveillance data;

[0006] Step 2: Reduce the dimension of the preprocessed data, then merge it into a data set, and divide the data set into a training set and a test set;

[0007] Step 3: Train a random forest model and an isolation forest model based on the data set, and combine the random forest model and the isolation forest model to generate an online model;

[0008] Step 4: Use the online model to conduct a risk assessment on the ammonia refrigeration system.

[0009] In the above risk assessment method for an ammonia refrigeration system, Step 1 includes the following specific steps:

[0010] Step 11: By judging the interval and data size at each time point, remove duplicate sensor data to ensure the uniqueness of each piece of data; use the interpolation method of autoregressive integrated moving average to process missing data and fill in the missing data;

[0011] Step 12: For traditional sensor data, perform Z-score standardization processing, and then extract basic statistical features. The basic statistical features are summarized into the data feature collection of the sensor;

[0012] Step 13: Use one-hot encoding for categorical data to obtain the one-hot encoding sets of categorical data for each sensor, and perform consistency checks using the data of multiple sensors. Calculate the Pearson correlation between sensors, eliminate the sensor data with low correlation, and retain the remaining one-hot encoding sets of categorical data;

[0013] Step 14: Process the gas concentration data and extract features to obtain the gas concentration data feature set;

[0014] Step 15: Process the video surveillance data and extract features to obtain the video surveillance data feature set.

[0015] In the above ammonia refrigeration system risk assessment method, step 2 includes the following specific steps:

[0016] Step 21: Perform principal component analysis dimensionality reduction on the data feature set, gas concentration data feature set, and video surveillance data feature set to obtain the main data source set, the first auxiliary data source set, and the second auxiliary data source set respectively;

[0017] Step 22: For the one-hot encoding set of categorical data, use the linear discriminant analysis method to find the linear combination that maximizes the distance between categories and minimizes the distance within categories, and form the categorical data feature vector set;

[0018] Step 23: Concatenate the main data source set, the first auxiliary data source set, the second auxiliary data source set, and the categorical data feature vector set to form a composite feature vector X combine :

[0019] X combine = [X temp , X pressure , X current ,....., X video ;

[0020] Step 24: The composite feature vector set constitutes a data set. The data set is divided into a training set and a test set. 80% of the data is the training set, and 20% of the data is the test set. The training set is used for training and verification.

[0021] In the above ammonia refrigeration system risk assessment method, in step 21, when performing principal component analysis dimensionality reduction, calculate the covariance matrix according to the data, then solve the eigenvalues and eigenvectors of the covariance matrix, and select the first k eigenvectors to form the principal components.

[0022] In the above ammonia refrigeration system risk assessment method, in step 22, for the one-hot encoding collection of categorical data, calculate the between-class scatter matrix and the within-class scatter matrix, then solve the generalized eigenvalue problem to obtain a set of eigenvectors, and select the eigenvectors that can maximize the ratio of between-class scatter to within-class scatter to form the categorical data eigenvector set.

[0023] In the above ammonia refrigeration system risk assessment method, step 3 includes the following specific steps:

[0024] Step 31: Train a random forest model based on the training set, and evaluate the performance of the model through the cross-validation method and the test set, using the accuracy A of the random forest model ram to comprehensively evaluate the model;

[0025] Step 32: Train an isolation forest model based on the training set and calculate the accuracy A of the isolation forest model according to the test set iso ;

[0026] Step 33: Assign weights w according to the random forest accuracy A ram and the isolation forest accuracy A iso ; if A ram > A iso , then the random forest weight w ram = 0.6, and the isolation forest weight w iso = 1 - w ram ; if A ram < A iso , then the random forest weight w ram = 0.4, and the isolation forest weight w iso = 1 - w ram ;

[0027] Combine the random forest model and the isolation forest model to generate an online model. The online model, based on the weights, fuses the prediction results of the random forest model and the isolation forest to output the final prediction result.

[0028] In the above ammonia refrigeration system risk assessment method, in step 33, during the operation of the online model, if any one of the models makes a misjudgment, its weight decreases by 0.02, and the weight value of the other model increases by 0.02.

[0029] In the above ammonia refrigeration system risk assessment method, in step 4, divide the risk level into high, medium, and low, correspond the risk level with the countermeasures, formulate a risk matrix, map the output result of the online model into the risk matrix, judge the current risk level of the ammonia refrigeration system, and execute corresponding measures according to the risk level.

[0030] In the above ammonia refrigeration system risk assessment method, the weights of each device and the device failure impact scores are predefined; the risk matrix formula is as follows:

[0031] R = F·I,

[0032] where R is the risk level, F is the occurrence frequency, and I is the impact degree; F can be converted through the output result P of the online model and the correction coefficient C; F = P×C;

[0033] The impact degree I = ∑(weight of the faulty device part * device failure impact score);

[0034] Based on the comprehensive analysis of the risk level, occurrence frequency, and impact degree, the risk level is judged.

[0035] In the above ammonia refrigeration system risk assessment method, the correction coefficient C is obtained from the device startup coefficient C run on the current day and the device failure coefficient C fau on the current day; C = C run ×C fau ;

[0036] C run = the number of devices started on the current day / the total number of devices;

[0037] C fau = 1 + 0.08×the number of device failures on the current day / the number of devices started on the current day, and 0.08 is the adjustment coefficient.

[0038] One of the technical solutions in the above technical solutions of the present invention has at least the following advantages or beneficial effects:

[0039] In the present invention, multiple data sources are integrated, and combined with the random forest model and the isolation forest model, a more comprehensive and accurate security assessment model is established. This model can obtain and analyze all available data in the computer room in real time to achieve a comprehensive perception and real-time monitoring of the system operation status, thereby improving the accuracy and timeliness of risk assessment, promptly discovering potential risks and issuing early warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present invention will be further described below with reference to the drawings and embodiments;

[0041] Figure 1 is a flowchart of the ammonia refrigeration system risk assessment method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0042] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0043] Ammonia itself also has certain safety hazards, especially its toxicity and flammability. Once a leakage or system failure occurs, it may pose a great threat to operators, the surrounding environment and equipment. Therefore, how to effectively ensure the safe operation of the ammonia refrigeration system has become an urgent problem to be solved in the current refrigeration field.

[0044] This application provides a risk assessment method for an ammonia refrigeration system. By integrating multiple data sources, a more comprehensive and accurate safety assessment model is established. This model can obtain and analyze all available data in the machine room in real time to achieve a comprehensive perception and real-time monitoring of the system operation state, thereby improving the accuracy and timeliness of risk assessment, promptly discovering potential risks and issuing early warnings.

[0045] Referring to Figure 1 As shown, a risk assessment method for an ammonia refrigeration system includes the following steps:

[0046] Step 1: Preprocess traditional sensor data, category data, gas concentration data, and video surveillance data;

[0047] The data preprocessing includes the following specific steps;

[0048] Step 11: Traditional sensor data, category data, gas concentration data, and video surveillance data are all data collected based on sensors. By judging the interval and data size at each time point, duplicate sensor data is removed to ensure the uniqueness of each piece of data;

[0049] The missing data is processed by the interpolation method of autoregressive integrated moving average (ARIMA) to fill in the missing data; assuming the missing data point is x t , based on the formula:

[0050] where φ1, φ2,.., φ p , are the parameters of the ARIMA model, ∈ t is the error term, is the predicted value of x t , and the predicted value is used to fill in the missing data;

[0051] For sensors with severely missing data, they can be selected for deletion and the data of these sensors will no longer be used.

[0052] Step 12: For traditional sensor data, perform Z-score standardization and then extract features;

[0053] Traditional sensor data is numerical data, as shown in Table 1:

[0054] Table 1 Traditional sensor data

[0055] Equipment Data type Ammonia storage tank Pressure, liquid level High-temperature compressor unit Current, load position, gas separation pressure, exhaust pressure Low-temperature compressor unit Current, load position, gas separation pressure, exhaust pressure Ethylene glycol unit Current, load position, gas separation pressure, exhaust pressure Barrel pump Pressure, liquid level Plate heat exchanger Pressure, liquid level, inlet water temperature, outlet water temperature Evaporative cooler Gas separation pressure, gas separation liquid level, receiver pressure, receiver liquid level

[0056] Each piece of data of the device is collected by at least one sensor. The data obtained by each sensor is a group. The data of the group is standardized based on the group. The Z-score standardization formula is as follows:

[0057] x i is the initial data, where μ i is the mean value of the data set, and σ is the standard deviation of the data set;

[0058] For |z i For data points with |>3, we judge them as outliers, and use the median of the data points to replace the outliers and keep records;

[0059] After standardization, the scale of the data has been unified, and there is no need to divide the data according to the sensor type. Standardization makes the data of different sensors on the same scale, reducing the impact caused by dimensional differences. Usually, these data can be processed uniformly, which accelerates subsequent model training, improves convergence speed and stability, improves model performance, and avoids certain features from having too much impact on the model.

[0060] The standardized data is z i (t), calculate basic statistical characteristics, including the following characteristics:

[0061] Mean

[0062] Standard Deviation

[0063] Skewness

[0064] Kurtosis

[0065] The basic statistical features are summarized into the data feature set of the sensor.

[0066] Step 13: Use one-hot encoding for the categorical data and use the data from multiple sensors to perform consistency checks, calculate the Pearson correlation between sensors, and remove sensor data with low correlation;

[0067] The category data includes the status of gas detection and alarm, etc. There are generally three states, including "normal", "fault", and "warning". One-hot encoding is performed to encode the status into binary features: normal [1, 0, 0], fault [0, 1, 0], warning [0, 0, 1], and the one-hot encoding set of the category data of each sensor is obtained;

[0068] After that, the category data is used to perform a consistency check using the data of multiple sensors, and the Pearson correlation p between the sensors is calculated ij ;

[0069]

[0070] where x ik 、x jk are the data points of sensors i and j respectively, are the data means of sensors i and j respectively;

[0071] If p ij is lower than the preset threshold, the data of the sensor with low correlation is removed, and the remaining one-hot encoding set of the category data is retained.

[0072] Step 14: Process the gas concentration data and extract features to obtain the gas concentration data feature set; the gas concentration data is obtained by gas detectors evenly distributed on the ceiling of the refrigeration system machine room. The gas concentration data provides numerical data on environmental safety, can reflect the dangerous situation in the environment, and provides supplementary information and enhances the analysis results for the main data source;

[0073] For the gas concentration data, the gas peak concentration, mean value, and fluctuation amplitude need to be calculated;

[0074] The fluctuation amplitude FF = max(x gas ) - min(x gas );

[0075] max(X gas ) is the gas peak concentration, and min(X gas ) is the minimum gas concentration; the gas peak concentration can reflect the extreme situation of gas leakage and help judge whether it exceeds the safety threshold; the mean value reflects the overall level of gas concentration and is used to evaluate the long-term exposure risk; the fluctuation amplitude is used to reflect the stability.

[0076] Step 15: Process the video surveillance data and extract features to obtain the video surveillance data feature set;

[0077] For video surveillance data, a convolutional neural network (CNN) is used to extract the image features of the device status in the picture; the feature sequence extracted by CNN is sent to the long short-term memory network (LSTM), combined with the unique hot encoding of the device status of each frame of the video to capture the dynamic characteristics of the device status changing over time; and a spatiotemporal joint feature that identifies the change of device status is generated;

[0078] The formula of CNN convolution layer is as follows: S=W*I+b, where S is the output feature map, W is the convolution kernel, I is the input image, and b is the bias.

[0079] Step 2: Reduce the dimension of the preprocessed data, merge them into a data set, and divide the data set into a training set and a test set.

[0080] Specifically, step 2 includes the following specific steps:

[0081] Step 21: Perform principal component analysis (PCA) dimensionality reduction on the data feature set, gas concentration data feature set, and video surveillance data feature set;

[0082] Take the data feature set X as an example, X=[x1,x2,....x i ];

[0083] First calculate the covariance matrix: μ is the mean;

[0084] Then solve the eigenvalues ​​and eigenvectors of the covariance matrix, select the first k eigenvectors to form the principal components and the main data source set; k is selected by calculating the cumulative variance contribution rate, and the first k eigenvectors are selected because the data corresponding to these eigenvectors have the largest variability and can most effectively represent the main information of the original data. The eigenvalues ​​and corresponding cumulative variance contribution rates of the first k eigenvectors should be high enough;

[0085] By analogy, the gas concentration data feature set can be reduced in dimension to a first auxiliary data source set, and the video surveillance data feature set can be reduced in dimension to a second auxiliary data source set;

[0086] The purpose of dimensionality reduction is to reduce the dimension of the data and remove redundant information, thereby improving the efficiency and generalization ability of the model. Through PCA, we can compress the original high-dimensional features into new low-dimensional features, while retaining the most important variability in the data as much as possible and reducing computational overhead.

[0087] Step 22: For the one-hot encoded set of category data, use the current discriminant analysis method (LDA) to find the linear combination that maximizes the distance between categories and minimizes the distance within categories;

[0088] Calculate the inter-class scatter matrix S B and the intra-class scatter matrix S w ;

[0089] Between-class scatter

[0090] Within-class scatter

[0091] where n i is the number of samples in the i-th class, and μ i is the mean of the i-th class, and μ is the mean of all data;

[0092] Then solve the generalized eigenvalue problem to obtain a set of eigenvectors (i.e., projection directions). These eigenvectors represent the main variation trends of the data in various directions. Select the eigenvectors that can maximize the ratio of the between-class scatter to the within-class scatter to form the class data eigenvector set.

[0093] Step 23: Concatenate the main data source set, the first auxiliary data source set, the second auxiliary data source set, and the class data eigenvector set to form a composite feature vector X combine :

[0094] X combine = [X temp , X pressure , X current ,....., X video ,

[0095] After feature fusion, if the feature dimension is too large, then execute Step 21 and Step 22 for dimensionality reduction.

[0096] Step 24: The composite feature vector collection constitutes a data set. Divide the data set into a training set and a test set. Generally, 80% of the data is the training set, and 20% of the data is the test set. The training set is used for training and validation.

[0097] Step 3: Perform offline model training according to the data set;

[0098] Step 31: Train a random forest model according to the training set, and evaluate the performance of the model through the cross-validation method and the test set;

[0099] Define T as the training set, and the prediction of the random forest is:

[0100]

[0101] where f k is the prediction of the k-th decision tree, and mode is the mode function;

[0102] To verify the accuracy of the model, use cross-validation (such as k-fold cross-validation) and the test set to calculate the accuracy A of the random forest model ramTo comprehensively evaluate the model.

[0103] The k-fold cross-validation formula is as follows:

[0104]

[0105] Where CV is the accuracy of cross-validation, and score(D i ) is the model score for the i-th fold.

[0106] Step 32: Train the Isolation Forest model according to the training set, use the Isolation Forest algorithm to perform anomaly detection on the data, recursively split the data points by randomly selecting a feature and setting a random split value until each point is "isolated", and the degree of isolation of the anomaly points is relatively high;

[0107] Define the data point x, and its anomaly score is calculated by the following formula:

[0108]

[0109] Where h(x) is the tree height of the isolated data point, and c(n) is a constant for normalizing the dataset size n;

[0110] To verify the accuracy of the model, calculate the accuracy A of the Isolation Forest model according to the test set iso .

[0111] Step 33: Allocate the weight w according to the Random Forest accuracy A ram and the Isolation Forest accuracy A iso . If A ram >A iso， , then the Random Forest weight w ram = 0.6, and the Isolation Forest weight w iso = 1 - w ram ; if A ram <A iso , then the Random Forest weight w ram = 0.4, and the Isolation Forest weight w iso = 1 - w ram . During the operation, if any model makes a misjudgment, the weight will decrease by 0.02, and the weight value of the other model will increase by 0.02;

[0112] Use the classification result of the Random Forest as the main basis for judging the safety state, and at the same time consider the anomaly data points detected by the Isolation Forest to generate an online model. Through the above weight calculation method, fuse the prediction results of the Random Forest and the Isolation Forest to obtain the final prediction result, that is, the probability value P of the failure of the refrigeration system.

[0113] Step 4: Use the online model to conduct risk assessment on the ammonia refrigeration system;

[0114] The risk levels are classified into high, medium, and low, corresponding to risk levels and countermeasures, and a risk matrix is formulated. The risk matrix formula is as follows:

[0115] R = F·I,

[0116] where R is the risk level, F is the occurrence frequency, and I is the impact degree;

[0117] F can be converted through the predicted probability value P output by the online model and the correction coefficient C;

[0118] C is obtained from the equipment startup coefficient C run and the equipment failure coefficient C fau The ammonia refrigeration system has a management background and can know which equipment has failed;

[0119] C run = the number of powered-on devices on the day / the total number of devices;

[0120] C fau = 1 + 0.08 × the number of equipment failures on the day / the number of powered-on devices on the day, and 0.08 is the adjustment coefficient;

[0121] C = C run ·C fau ,F = P × C run × C fau ;

[0122] The impact degree I = ∑(weight of the failed equipment part * failure impact score). The equipment weights are shown in Table 2, and the equipment failure impact scores are shown in Table 3;

[0123] Table 2 Equipment Weights

[0124] Equipment type Weight Description Ammonia storage tank 1.0 High-risk equipment, with a very large impact (safety issue) High-temperature unit 0.9 Key equipment, with a relatively large impact on failures Low-temperature unit 0.85 Key equipment, with a relatively large impact on failures Ethylene glycol unit 0.8 Key equipment, with a relatively large impact on failures Barrel pump 0.75 Crucial for the circulation system, with a relatively large impact Plate heat exchanger 0.6 Has a relatively small impact on the system, but is still important Evaporative cooler 0.4 Auxiliary equipment, with a relatively small impact

[0125] Table 3 Equipment Failure Impact Scores

[0126]

[0127] Map the output structure of the online model into the risk matrix as shown in Table 4, judge the current risk level of the ammonia refrigeration system, and implement corresponding measures according to the risk level.

[0128] Table 4 Risk Matrix

[0129]

[0130] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A risk assessment method for an ammonia refrigeration system, characterized in that, The following steps are involved: Step 1: Preprocess traditional sensor data, category data, gas concentration data, and video surveillance data; Step 2: Reduce the dimension of the preprocessed data, merge them into a data set, and divide the data set into a training set and a test set; Step 3: Perform random forest model training and isolation forest model training according to the data set, and combine the random forest model and isolation forest model to generate an online model; Step 4: Conduct risk assessment of the ammonia refrigeration system using the online model.

2. The risk assessment method for an ammonia refrigeration system according to claim 1, characterized in that Step 1 includes the following specific steps: Step 11: By determining the interval and data size at each time point, duplicate sensor data is removed to ensure that each piece of data is unique; the missing data is processed using the autoregressive integral moving average interpolation method to fill in the missing data; Step 12: For traditional sensor data, perform Z-score standardization processing, and then extract basic statistical features, which are summarized into the data feature set of the sensor; Step 13: Use one-hot encoding for the category data to obtain the one-hot encoding set of the category data of each sensor, and use the data of multiple sensors to perform consistency check, calculate the Pearson correlation between sensors, eliminate sensor data with low correlation, and retain the remaining one-hot encoding set of category data; Step 14: Process the gas concentration data and extract features to obtain a gas concentration data feature set; Step 15: Process the video surveillance data and extract features to obtain a video surveillance data feature set.

3. The risk assessment method for an ammonia refrigeration system according to claim 2, wherein Step 2 includes the following specific steps: Step 21: Perform principal component analysis and dimensionality reduction on the data feature set, the gas concentration data feature set, and the video monitoring data feature set to obtain a main data source set, a first auxiliary data source set, and a second auxiliary data source set, respectively; Step 22: For the one-hot encoding collection of category data, use the current discriminant analysis method to find the linear combination that maximizes the distance between categories and minimizes the distance within categories, and form a feature vector set of category data; Step 23: Concatenate the primary data source set, the first auxiliary data source set, the second auxiliary data source set, and the category data feature vector set to form a composite feature vector X combine : X combine = [X temp , X pressure , X current ,....., X video ; Step 24: The composite feature vector collection constitutes a data set, and the data set is divided into a training set and a test set. 80% of the data is the training set and 20% of the data is the test set. The training set is used for training and verification.

4. The risk assessment method for an ammonia refrigeration system according to claim 3, wherein In step 21, when performing principal component analysis dimensionality reduction, the covariance matrix is ​​calculated based on the data, and then the eigenvalues ​​and eigenvectors of the covariance matrix are solved, and the first k eigenvectors are selected to form the principal components.

5. The risk assessment method for an ammonia refrigeration system according to claim 3, characterized in that, In step 22, for the one-hot encoded set of category data, the inter-class scatter matrix and the intra-class scatter matrix are calculated, and then the generalized eigenvalue problem is solved to obtain a set of eigenvectors, and the eigenvector that can maximize the ratio of the inter-class scatter to the intra-class scatter is selected to form a categorical data feature vector set.

6. The risk assessment method for an ammonia refrigeration system according to claim 1, characterized in that, Step 3 includes the following specific steps: Step 31: Train a random forest model based on the training set, and evaluate the performance of the model by means of cross-validation method and the test set, using the accuracy A of the random forest model ram to comprehensively evaluate the model; Step 32: Train the Isolation Forest model based on the training set and calculate the accuracy A of the Isolation Forest model according to the test set iso ; Step 33: According to the random forest accuracy rate A ram and the isolation forest accuracy rate A iso allocate the weight w; if A ram >A iso , then the random forest weight w ram = 0.6, and the isolation forest weight w iso = 1 - w ram ; if A ram <A iso , then the random forest weight w ram = 0.4, and the isolation forest weight w iso = 1 - w ram ; The random forest model and the isolation forest model are combined to generate an online model. The online model fuses the prediction results of the random forest model and the isolation forest model based on weights to output the final prediction results.

7. The risk assessment method for an ammonia refrigeration system according to claim 6, wherein In step 33, when the online model is running, if any model makes a misjudgment, its weight decreases by 0.02, and the weight of the other model increases by 0.

02.

8. The risk assessment method for an ammonia refrigeration system according to claim 1, characterized in that, In step 4, the risk levels are divided into high, medium, and low, and the risk levels are corresponded to the countermeasures, and a risk matrix is formulated. The output result of the online model is mapped into the risk matrix to judge the current risk level of the ammonia refrigeration system, and corresponding measures are implemented according to the risk level.

9. The risk assessment method for an ammonia refrigeration system according to claim 8, characterized in that, The weights of each device and the device failure impact scores are predefined; the risk matrix formula is as follows: R = F·I, where R is the risk level, F is the occurrence frequency, and I is the impact degree; F can be converted by the output result P of the online model and the correction coefficient C; F = P×C; The impact degree I = ∑(weight of the faulty device part * failure impact score); Based on the comprehensive analysis of the risk level, occurrence frequency, and impact degree, the risk level is judged.

10. The risk assessment method for an ammonia refrigeration system according to claim 9, characterized in that, The correction coefficient C is obtained from the equipment startup coefficient C on the same day run and the equipment failure coefficient C on the same day fau ; C = C run × C fau ; C run = Number of devices powered on on the day / Total number of devices; C fau = 1 + 0.08 × (number of equipment failures on the day / number of powered-on equipment on the day), where 0.08 is the adjustment coefficient.

Images