Compressor operation status monitoring method and system based on medical oxygen production system

Abstract

The present invention relates to the field of data processing technology, and the present invention relates to a compressor operation status monitoring method and system based on a medical oxygen production system, wherein the method comprises: obtaining intermediate results corresponding to each participating node; screening out honest nodes among all participating nodes; taking the intermediate results corresponding to the honest nodes as expected results, calculating the outliers of each expected result; determining the expected results with outliers greater than a threshold as target results; performing model aggregation through all target results to obtain a main model and publishing the main model to a blockchain, so that each participating node obtains the main model through the blockchain. The present invention avoids the decrease in the accuracy of the main model caused by attacks on centralized nodes by encrypting the training intermediate results of the participating nodes and publishing them to the blockchain for aggregation.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a method and system for monitoring the operating status of a compressor based on a medical oxygen production system. Background Art

[0002] With the rapid progress in the medical field, hospitals have an increasing demand for medical oxygen compressors, which are essential for maintaining patient safety. In order to protect data security, hospitals generally use local area networks to manage key medical equipment, which leads to the lack of interoperability of equipment data. Due to the differences in resources and technology applications between hospitals, the lack of data sharing has seriously affected the efficiency and consistency of medical equipment management. Therefore, the lack of effective cross-hospital data sharing technology has become a major obstacle to improving the quality of medical services and equipment stability.

[0003] In order to protect the operating data of medical devices and optimize the model used to predict the operating status of medical devices using the operating data of medical devices, the existing technology usually uses federated learning to enable each participating node to optimize the same initial model locally through the corresponding medical device operating data, and pass the intermediate data of the local model obtained after training (usually the parameters of the local model) to the trusted node, which performs model aggregation to obtain a main model with higher accuracy.

[0004] The traditional federated learning framework requires a centralized node to aggregate and inherit the submitted local training results and obtain a global model (main model). For example, the Chinese patent application document with publication number CN117196069A discloses a federated learning method, which includes: signing a federated learning cooperation agreement, which includes the federated learning permissions and usage rules of multiple parties; establishing a federated learning computing environment according to the cooperation agreement; each data participant prepares federated learning training data locally according to unified data processing rules; and starting a federated learning development task to conduct a federated learning process. However, the security of FL (Federated Learning) depends entirely on the security of centralized nodes (trusted nodes). Once the centralized nodes are hacked, the entire FL system will fail. And because there are many nodes participating in FL, it is impossible to guarantee that all participating nodes are honest nodes (non-malicious nodes). Because the original data cannot leave the local participating nodes, it is difficult to monitor the original data from malicious nodes for poisoning attacks. When the data of malicious nodes is used to update the global model, the accuracy of the obtained global model may be poor. Summary of the invention

[0005] In order to solve the above-mentioned technical problems that the security of centralized nodes cannot be guaranteed and all participating nodes are honest nodes, and when the data of malicious nodes is used when updating the global model, the accuracy of the obtained global model may be poor, the present invention provides solutions in the following aspects.

[0006] In a first aspect, a method for monitoring the operating status of a compressor in a medical oxygen production system includes: obtaining an intermediate result of an operating status prediction model corresponding to a local compressor of each participating node on the chain, wherein the intermediate result includes the trained operating status prediction model. M parameters; filter out the honest nodes among all participating nodes, take the intermediate results corresponding to the honest nodes as the expected results, and calculate the k The cluster value of the expected result m k , , For the i The expected result j The value of the parameter, For the k The expected result j The value of the parameter, For all expected results, j The standard deviation of the parameter, j , i are all positive integers, N is the number of expected results, exp() is e An exponential function with a base of ; determining the expected result with a cluster value greater than a threshold as the target result, performing model aggregation through all target results to obtain the main model and publishing the main model to the blockchain, so that each participating node can obtain the main model through the blockchain.

[0007] The beneficial effects are as follows: on the one hand, the present invention aggregates the intermediate results of the trained models of each participating node on the chain, avoiding the situation where the trusted node (centralized node) is attacked and the accuracy of the aggregated main model is low. Furthermore, after the present invention publishes the trained main model to the blockchain, each participating node obtains the main model through the blockchain to optimize the main model through local data, and aggregates the optimized model again on the chain. By iterating this step, the accuracy of the model obtained by model aggregation is continuously increased. On the other hand, the present invention sequentially screens the nodes and the intermediate results of the nodes to obtain the most reliable intermediate results (target results) to participate in model aggregation, so that the accuracy of the final main model is higher.

[0008] Preferably, obtaining the intermediate result corresponding to any participating node includes: for any participating node, obtaining the ciphertext corresponding to the participating node through the blockchain, and decrypting the ciphertext through the public key to obtain the intermediate result corresponding to the participating node, wherein any participating node encrypts the intermediate result corresponding to the participating node through the public key to obtain the ciphertext, and publishes the ciphertext to the blockchain.

[0009] The present invention enables participating nodes to encrypt intermediate results and publish them on the blockchain. Only nodes holding corresponding public keys can decrypt these data, thereby increasing data security and preventing data leakage.

[0010] Preferably, the honest nodes among all participating nodes are screened through secure multi-party computing, and the screening process includes: distributing a predetermined computing share to each participating node through the blockchain, wherein each participating node performs calculations according to a preset addition gate and / or multiplication gate and generates a calculation result and publishes it to the blockchain; obtaining the calculation result of each participating node through the blockchain; for any participating node, judging whether the participating node has tampered with the share according to the calculation result of the participating node; and determining that the participating node whose share has not been tampered with is an honest node.

[0011] The present invention screens out honest nodes among the participating nodes through secure multi-party computing, and only allows the intermediate results corresponding to the honest nodes to participate in the model aggregation process. Based on this, it avoids the situation where malicious nodes deliberately tamper with the intermediate results so that the main model obtained by model aggregation has low accuracy.

[0012] Preferably, judging whether the participating node has tampered with the share according to the calculation result of the participating node includes: in response to the computing resource consumption of the participating node not matching its share, determining that the participating node has tampered with the share; calculating the target ratio corresponding to the participating node α : , a 1 The same number of bytes, a 2 is the number of bytes of the calculation result corresponding to the participating node, where the same number of bytes is calculated a 1 The method comprises: comparing the calculation result corresponding to the participating node with the preset standard result byte by byte, and determining the number of bytes that are the same between the two as the same byte number. a 1 ; In response to the target ratio corresponding to the participating node α A ratio threshold that is less than or equal to a preset size determines the tampering share of the participating node.

[0013] Preferably, the compressor operation status monitoring method based on the medical oxygen production system further includes correcting the cluster value of each expected result before determining the expected result with a cluster value greater than a threshold as the target result, wherein the cluster value of the first expected result is corrected. k The formula for correcting the cluster value of the expected result is: m' k = m k × L k , among which for the k Expected results, m k is the cluster value before correction, m' k is the corrected cluster value, correction coefficient L k satisfy:

[0014] .

[0015] in For all expected results, s The maximum value among the parameters, For all expected results, s The minimum value among the parameters, norm () is the normalization function, s Is a positive integer.

[0016] Preferably, the compressor operating status monitoring method based on the medical oxygen production system also includes screening the trained models of the participating nodes corresponding to each target result before model aggregation through all target results: constructing a verification set, the verification set includes a predetermined number of compressor operating data subsets, wherein any operating data subset includes the operating data of the industrial computer at a moment and a label of whether the compressor fails at this moment; for the trained model of any participating node, inputting the verification set into the model to obtain multiple prediction results; comparing each prediction result with the label of its corresponding operating data subset, and in response to the number of operating data subsets with the same prediction result and label being greater than a preset quantity threshold, determining that the model participates in model aggregation.

[0017] The present invention verifies the target result through the verification set to avoid overfitting of the final main model, that is, to prevent the main model from learning the noise and specific models in the training data instead of features with strong generalization ability.

[0018] Preferably, the step of performing model aggregation through all target results to obtain the main model comprises: performing weighted summation on the nodes corresponding to all target results to obtain the parameters of the main model, wherein for the first h Parameters ah , which satisfies: , is the first weight of the preset size, is the second weight of the preset size, For the preset size X -1 weight, For the preset size X Weight, The first target result h The value of the parameter, The second target result h The value of the parameter, For the X -1 the value of the hth parameter of the target result, For the X The first target result h The value of the parameter, X is the number of target results; a main model is constructed according to the parameters of the main model, wherein the architecture of the main model is the same as the architecture of the trained model corresponding to each participating node.

[0019] In a second aspect, the memory stores a computer program, and the processor executes the computer program to implement the compressor operation status monitoring method based on a medical oxygen production system as described in any one of the above invention contents.

[0020] The beneficial effects of the present invention are as follows: the present invention avoids the accuracy of the main model being reduced due to attacks on centralized nodes by encrypting the intermediate training results of participating nodes and publishing them to the blockchain for aggregation, while ensuring that only nodes holding corresponding keys can decrypt data, thereby enhancing security. By screening honest nodes to participate in model aggregation, malicious tampering is prevented and the accuracy of the main model is ensured. In addition, the model is verified using a validation set to avoid overfitting, thereby improving the generalization ability, stability and security of the model and ensuring a more efficient and reliable distributed model training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0022] Figure 1 is a flowchart schematically illustrating the steps of a method for monitoring the operating status of a compressor based on a medical oxygen production system according to an embodiment of the present invention;

[0023] Figure 2Schematically shows a structural block diagram of a compressor operation status monitoring system based on a medical oxygen production system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0025] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0026] The present invention designs the smart contract to receive data input (i.e., the smart contract can obtain the intermediate results of each participating node), and then calculates the cluster value of each intermediate result in the contract (i.e., on-chain), screens out honest nodes from all participating nodes, and performs model aggregation.

[0027] Figure 1 The figure is a flowchart schematically showing the steps of a method for monitoring the operating status of a compressor in a medical oxygen production system according to an embodiment of the present invention.

[0028] like Figure 1 As shown, the compressor operation status monitoring method based on the medical oxygen production system includes steps S1 to S3.

[0029] Step S1: Obtain the intermediate results of the operation status prediction model corresponding to the local compressor of each participating node on the chain.

[0030] The intermediate results include the trained operation status prediction model M parameters;

[0031] Any participating node obtains the model to be trained through the blockchain, and the participating node trains the model to be trained through the operation data of the local compressor to obtain a trained model. M The parameter is the intermediate result corresponding to the participating node. M Is a positive integer; filter out honest nodes from all participating nodes.

[0032] It should be noted that, at the beginning, each participating node is a client that collects the compressor operation data of the medical oxygen production system. Each participating node is configured with a model to be trained with the same architecture and the same initialization weights, and each participating node trains the model to be trained using the local compressor operation data. Among them, each participating node trains the model to be trained using the same learning rate and loss model through gradient descent.

[0033] In subsequent iterations, each participating node obtains the main model from the blockchain as the new model to be trained, and uses the same local compressor operation data to train the new training model to obtain a new trained model.

[0034] In one embodiment, obtaining the intermediate result corresponding to any participating node includes: for any participating node, obtaining the ciphertext corresponding to the participating node through the blockchain, and decrypting the ciphertext through the public key to obtain the intermediate result corresponding to the participating node, wherein any participating node encrypts the intermediate result corresponding to the participating node through the public key to obtain the ciphertext, and publishes the ciphertext to the blockchain.

[0035] In one embodiment, a preset initial model is published so that each participating node obtains the initial model, and the architecture of the initial model is the same as the architecture of the main model; for any participating node, the participating node builds a training set based on the historical operation data of the local compressor, and the historical operation data includes the temperature, pressure, current, voltage of the compressor of the participating node at multiple times and a label of whether the compressor is faulty. The participating node trains the initial model through the training set to obtain the operating status prediction model of the participating node.

[0036] Step S2: Take the intermediate result corresponding to the honest node as the expected result and calculate the cluster value of the expected result.

[0037] In one embodiment, screening out honest nodes from all participating nodes through secure multi-party computing includes: distributing a predetermined computing share to each participating node, wherein each participating node performs computing according to a preset addition gate and / or multiplication gate and generates a computing result and publishes it to the blockchain; obtaining the computing result of each participating node; for any participating node, judging whether the participating node has tampered with the share according to the computing result of the participating node; and determining that the participating node whose share has not been tampered with is an honest node.

[0038] Further, judging whether the participating node has tampered with the share according to the calculation result of the participating node includes: in response to the computing resource consumption of the participating node not matching its share, determining that the participating node has tampered with the share; calculating the target ratio corresponding to the participating node α : , a 1 The same number of bytes, a 2 is the number of bytes of the calculation result corresponding to the participating node, where the same number of bytes is calculated a 1 The method comprises: comparing the calculation result corresponding to the participating node with the preset standard result byte by byte, and determining the number of bytes that are the same between the two as the same byte number. a 1; In response to the target ratio corresponding to the participating node α A ratio threshold that is less than or equal to a preset size determines the tampering share of the participating node.

[0039] Among them, calculate the k The cluster value of the expected result m k The formula is:

[0040] .

[0041] For the i The expected result j The value of the parameter, For the k The expected result j The value of the parameter, For all expected results, j The standard deviation of the parameter, j , i are all positive integers, N is the number of expected results, exp() is e An exponential function with a base of ; the expected result of determining the cluster value greater than the threshold is the target result.

[0042] It should be noted that for k The expected result j parameters, Indicates the outlier value of the parameter. The larger the outlier value of the parameter, the greater the probability that the parameter is abnormal data. At this time, the cluster value of the parameter m k The smaller, Indicates all expected results j The average value of a parameter, the difference between the value of the parameter and the average value (the difference is ), the larger the outlier value of the parameter. Further, through all the expected results corresponding to the j The standard deviation of the parameter Correct the parameter. Under normal circumstances (i.e. the model corresponding to the expected result is normal data suitable for improving the accuracy of the main model), the first j The difference between the first parameter and other parameters is large, so the j The standard deviation of the parameter The larger the value, the higher the expected result. j The smaller the outlier value of the parameter is.

[0043] Determine k The sum of the outliers of all parameters of the desired results is the outlier value of the expected result. The larger the outlier value of the expected result (the cluster value of the parameter m k The smaller the expected result is, the greater the probability that the expected result is an abnormal result, that is, the smaller the accuracy of the model corresponding to the expected result is, and the less suitable it is for model aggregation. Based on this, the larger the cluster value of the expected result is, the more suitable the expected result is for model aggregation.

[0044] Furthermore, the compressor operation status monitoring method based on the medical oxygen production system also includes correcting the cluster value of each expected result before determining the expected result with a cluster value greater than a threshold as the target result, wherein the cluster value of the first expected result is corrected. k The formula for correcting the cluster value of the expected result is: m' k = m k × L k , among which for the k Expected results, m k is the cluster value before correction, m' k is the corrected cluster value, correction coefficient L k satisfy:

[0045] .

[0046] in For all expected results, s The maximum value among the parameters, For all expected results, s The minimum value among the parameters, norm () is the normalization function, s Is a positive integer.

[0047] It should be noted that The larger the k The expected result s The parameters correspond to all the expected results. s The greater the difference between the maximum values ​​of the parameters, the k The expected result s The less likely the parameter is to be abnormal data (i.e., influence), the k The larger the correction factor of the expected result, the greater the The larger the k The expected result s The parameters correspond to all the expected results. s The greater the difference between the minimum values ​​of the parameters, the kThe expected result s The less likely the parameter is to be abnormal data (i.e., influence), the k The larger the correction factor for the expected result.

[0048] Step S3: Determine the expected result with a cluster value greater than a threshold as the target result, perform model aggregation through all target results to obtain a main model and publish the main model to the blockchain, so that each participating node can obtain the main model through the blockchain.

[0049] In one embodiment, weighted summation is performed on all nodes corresponding to the target results to obtain the parameters of the main model, where for the first h Parameters a h , which satisfies:

[0050] .

[0051] is the first weight of the preset size, is the second weight of the preset size, For the preset size X -1 weight, For the preset size X Weight, The first target result h The value of the parameter, The second target result h The value of the parameter, For the X -1 the value of the hth parameter of the target result, For the X The first target result h The value of the parameter, X is the number of target results; a main model is constructed according to the parameters of the main model, wherein the architecture of the main model is the same as the architecture of the trained model corresponding to each participating node.

[0052] Furthermore, the present invention also includes screening the trained models of the participating nodes corresponding to each target result before performing model aggregation through all target results: constructing a verification set, the verification set including a predetermined number of operating data subsets of compressors, wherein any operating data subset includes the operating data of the industrial computer at a moment and a label of whether the compressor fails at that moment; for the trained model of any participating node, inputting the verification set into the model to obtain multiple prediction results; comparing each prediction result with the label of its corresponding operating data subset, and in response to the number of operating data subsets with the same prediction result and label being greater than a preset quantity threshold, determining that the model participates in model aggregation.

[0053] The validation set is a data set used for model evaluation and parameter adjustment in machine learning and deep learning. The validation set is used to evaluate and optimize the performance of the model during the training process. During the training process, the validation set can help detect whether overfitting has occurred. Generally, the error on the training set will decrease as the training progresses, while the error on the validation set may begin to increase at some point. By monitoring the performance on the validation set, training can be stopped early when overfitting occurs (this technique is called early stopping).

[0054] Furthermore, the present invention also includes repeated training for multiple times to obtain an updated main model. The step of repeated training once includes: obtaining the optimized main model of each participating node, wherein after each participating node obtains the main model, the main model is optimized and trained through the operation data of the local compressor to obtain the optimized main model; honest nodes are screened out from the participating nodes, and nodes with larger cluster values ​​of the intermediate results of the optimized main model corresponding to the honest nodes are screened out, and such nodes are determined as target nodes; model aggregation is performed through the intermediate results of the optimized main model corresponding to the target node to obtain the updated main model, and the main model is published so that each participating node obtains the updated main model.

[0055] In one embodiment, in response to the number of repeated trainings reaching a predetermined number, the updated main model at the current moment is confirmed as the target main model, and the target main model is published so that each participating node obtains the target main model.

[0056] It should be noted that after the main model is published to the blockchain, each participating node can access the blockchain to obtain the main model, and each participating node can optimize the main model through the local compressor operation data, and after the optimization is completed, the intermediate results of the model (model parameters) are aggregated on the chain through smart contracts to achieve iteration.

[0057] Among them, each participating node obtains the target master model through the blockchain and predicts the local compressor operating status through the target master model.

[0058] It should be noted that all participating nodes can obtain the target master model by accessing the blockchain and predict the operating status of the compressor through the master model. In one embodiment, the target master model, the master model, and the trained models corresponding to each participating node are used to predict whether the compressor fails. The operating data of the compressor corresponding to each participating node is configured with a label, which is 0 or 1. When the label of the operating data of the compressor corresponding to a participating node is 1, when the operating data of the compressor corresponding to the participating node is collected, the air compressor operates normally; when the label of the operating data of the compressor corresponding to a participating node is 0, when the operating data of the compressor corresponding to the participating node is collected, the air compressor fails.

[0059] Figure 2 FIG. 4 is a block diagram schematically showing a structure of a compressor operation status monitoring system based on a medical oxygen production system according to the present embodiment.

[0060] The present invention also provides a compressor operation status monitoring system based on a medical oxygen production system. Figure 2 As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the compressor operation status monitoring method based on the medical oxygen production system according to the first aspect of the present invention is implemented.

[0061] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0062] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0063] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for monitoring the operating status of a compressor in a medical oxygen production system, characterized in that: include: Obtain the intermediate results of the operation status prediction model corresponding to the local compressor of each participating node on the chain, where the intermediate results include the trained operation status prediction model M parameters; Filter out the honest nodes among all participating nodes, take the intermediate results corresponding to the honest nodes as the expected results, and calculate the k The cluster value of the expected result m k , , For the i The expected result j The value of the parameter, For the k The expected result j The value of the parameter, For all expected results, j The standard deviation of the parameter, j , i are all positive integers, N is the number of expected results, exp() is e An exponential function with base ; The expected result with a cluster value greater than a threshold is determined as the target result, and the model is aggregated through all the target results to obtain the main model and publish the main model to the blockchain, so that each participating node can obtain the main model through the blockchain.

2. The method for monitoring the operating status of a compressor based on a medical oxygen production system according to claim 1, characterized in that: Obtaining the intermediate results corresponding to any participating node includes: For any participating node, the ciphertext corresponding to the participating node is obtained through the blockchain, and the ciphertext is decrypted through the public key to obtain the intermediate result corresponding to the participating node, wherein any participating node encrypts the intermediate result corresponding to the participating node through the public key to obtain the ciphertext, and publishes the ciphertext to the blockchain.

3. The method for monitoring the operating status of a compressor based on a medical oxygen production system according to claim 1, characterized in that: The honest nodes among all participating nodes are screened through secure multi-party computing. The screening process includes: Distribute the predetermined computing share to each participating node, wherein each participating node performs the calculation according to the preset addition gate and / or multiplication gate and publishes the generated calculation result to the blockchain; Obtain the calculation results of each participating node; For any participating node, determine whether the participating node has tampered with the share based on the calculation result of the participating node; Participating nodes that have not tampered with their shares are determined to be honest nodes.

4. The method for monitoring the compressor operation status based on the medical oxygen production system according to claim 3 is characterized in that: Judging whether the participating node has tampered with the share according to the calculation result of the participating node includes: In response to the computing resource consumption of the participating node not matching its share, determining that the participating node tampered with the share; Calculate the target ratio corresponding to the participating node α : , a 1 is the same number of bytes, a 2 is the number of bytes of the calculation result corresponding to the participating node, where the same number of bytes is calculated a 1 includes: comparing the calculation result corresponding to the participating node with the preset standard result byte by byte, and determining the number of bytes that are the same between the two as the same byte number a 1; in response to the target ratio corresponding to the participating node α A ratio threshold that is less than or equal to a preset size determines the tampering share of the participating node.

5. The method for monitoring the compressor operation status based on the medical oxygen production system according to claim 1, characterized in that: The method further includes modifying the cluster value of each expected result before determining the expected result with a cluster value greater than a threshold as the target result, wherein the cluster value of the first expected result is modified. k The formula for correcting the cluster value of the expected result is: m' k = m k × L k , among which for the k Expected results, m k is the cluster value before correction, m' k is the corrected cluster value, correction coefficient L k satisfy: ； in For all expected results, s The maximum value among the parameters, For all expected results, s The minimum value among the parameters, norm () is the normalization function, s Is a positive integer.

6. The method for monitoring the operating status of a compressor based on a medical oxygen production system according to claim 1, characterized in that: It also includes screening the trained models of the participating nodes corresponding to each target result before aggregating the models through all target results: Constructing a validation set, wherein the validation set includes a predetermined number of operating data subsets of compressors, wherein any operating data subset includes operating data of the industrial computer at a time and a label indicating whether the compressor fails at the time; For a trained model of any participating node, a validation set is input into the model to obtain multiple prediction results; each prediction result is compared with the label of its corresponding running data subset, and in response to the number of running data subsets with the same prediction results and labels being greater than a preset quantity threshold, the model is determined to participate in model aggregation.

7. The method for monitoring the operating status of a compressor based on a medical oxygen production system according to claim 1, characterized in that: The method of performing model aggregation through all target results to obtain a main model includes: The weighted sum of all nodes corresponding to the target results is performed to obtain the parameters of the main model, where for the first h Parameters a h , which satisfies: , is the first weight of the preset size, is the second weight of the preset size, The preset size X -1 weight, The preset size X Weight, The first target result h The value of the parameter, The second target result h The value of the parameter, For the X -1 the value of the hth parameter of the target result, For the X The target result h The value of the parameter, X is the number of target results; a main model is constructed according to the parameters of the main model, wherein the architecture of the main model is the same as the architecture of the trained model corresponding to each participating node.

8. The method for monitoring the operating status of a compressor based on a medical oxygen production system according to claim 7, characterized in that: Also includes: Publishing a preset initial model so that each participating node obtains the initial model, wherein the architecture of the initial model is the same as that of the main model; For any participating node, the participating node constructs a training set based on the historical operating data of the local compressor, wherein the historical operating data includes the temperature, pressure, current, voltage of the compressor of the participating node at multiple times and a label indicating whether the compressor is faulty. The participating node trains the initial model through the training set to obtain the operating status prediction model of the participating node.

9. A compressor operation status monitoring system based on a medical oxygen production system, comprising a processor and a memory, wherein the memory stores a computer program, characterized in that: The processor executes the computer program to implement the compressor operating status monitoring method based on the medical oxygen production system according to any one of claims 1 to 8.

Images