Gas cylinder management method and system

Abstract

The invention discloses a gas cylinder management method and system, and the method comprises the steps: obtaining time sequence data, material features and maintenance records from gas cylinders of each material, building a gas cylinder life prediction task based on the time sequence data, the material features and the maintenance records by employing a PC algorithm, and carrying out the management of the gas cylinder life according to the built gas cylinder life prediction task. Related characteristics influencing the service life of the gas cylinder are screened out; the related features are sent into a graph convolutional neural network model for fusion, knowledge migration is carried out according to the number of the fused features, and a life prediction result of the gas cylinder of each material is obtained; the knowledge migration effect is evaluated and optimized, so that the accuracy of predicting the service life of the gas cylinder made of each material is improved, and the intelligentization of gas cylinder management is realized; according to the method, deep knowledge transmission from the gas cylinders with rich data to the gas cylinders made of rare materials is realized by adopting the meta-learning model and the deep residual network, the gas cylinders are uniformly managed by outputting gas cylinder service life prediction, and a basis is provided for intelligent gas cylinder management.

Description

Technical Field

[0001] The present invention relates to the technical field of gas cylinder management, and particularly to a gas cylinder management method and system. Background Art

[0002] Gas cylinder management, as a key area of industrial safety and environmental protection, has witnessed a remarkable evolution from traditional manual records to modern intelligent management. In the early days, gas cylinder management mainly relied on manual inspections and paper-based maintenance records to evaluate the safety status by regularly checking the appearance, pressure, and service life of gas cylinders. With the acceleration of industrialization and the rise of Internet of Things (IoT) technology, data acquisition and remote monitoring systems based on sensors have gradually become the mainstream trend in gas cylinder management. For example, by installing pressure sensors and temperature sensors on gas cylinders, real-time monitoring of time-series data can be achieved, and data storage and preliminary analysis can be carried out in combination with a cloud computing platform, thereby improving the automation level of management. In addition, statistical models and empirical formulas are also widely used in gas cylinder life prediction, attempting to infer the aging law of gas cylinders through historical data.

[0003] Although the existing technologies have made great progress in the automation and digitalization of gas cylinder management, their deficiencies are still significant, especially in the accuracy and adaptability of life prediction. Firstly, traditional life prediction models are mostly based on statistical regression or fixed empirical formulas, ignoring the complex interaction between the material properties of gas cylinders (such as metal fatigue and corrosion resistance) and the use environment (such as temperature, humidity, and maintenance frequency), resulting in limited generalization ability of prediction results and difficulty in adapting to the attenuation laws of gas cylinders with different materials. Secondly, when dealing with multi-source heterogeneous data, existing methods usually only adopt feature splicing or dimensionality reduction processing, failing to effectively capture the deep correlation between data, which limits the expression ability of the prediction model. In addition, for gas cylinders made of rare materials with scarce data, traditional transfer learning methods mostly stay at the level of shallow feature transfer, lacking a cross-domain transfer mechanism for deep knowledge, resulting in insufficient robustness and accuracy of the model in small-sample scenarios. These deficiencies jointly restrict the further development of gas cylinder management towards intelligence and high precision. Therefore, there is an urgent need for a method to break through the limitations of existing technologies and promote the overall optimization of gas cylinder life prediction and management. Summary of the Invention

[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a gas cylinder management method to solve the problems mentioned in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a gas cylinder management method, including:

[0008] Obtain time series data, material characteristics, and maintenance records from gas cylinders of each material. Based on the time series data, material characteristics, and maintenance records, use the PC algorithm to establish a gas cylinder life prediction task, and according to the established gas cylinder life prediction task, screen out relevant characteristics that affect the gas cylinder life;

[0009] Send the relevant characteristics into a graph convolutional neural network model for fusion, and perform knowledge transfer according to the number of fused characteristics to obtain the life prediction results for gas cylinders of each material;

[0010] By evaluating and optimizing the knowledge transfer effect, the accuracy of the life prediction of gas cylinders of each material is improved, and the intelligent management of gas cylinders is realized.

[0011] As a preferred solution of the gas cylinder management method of the present invention, wherein: using the PC algorithm, using the PC algorithm, establishing a gas cylinder life prediction task, and according to the established gas cylinder life prediction task, screening out relevant characteristics that affect the gas cylinder life, including:

[0012] Define different basic acquisition characteristics for gas cylinders of each material to form a data set D of different basic acquisition characteristics under each material of gas cylinders;

[0013] Through the PC algorithm, infer the causal relationship between the basic acquisition characteristics under each material of gas cylinders and their corresponding gas cylinder lives from the data set D, and generate a directed acyclic graph;

[0014] For each basic acquisition characteristic, calculate its average causal effect value and mutual information value corresponding to the gas cylinder life;

[0015] Define an importance score through the average causal effect value and the mutual information value, and select the top k basic acquisition characteristics with the highest importance score;

[0016] Wherein, k is determined by the total number of basic acquisition characteristics and the number of gas cylinders of each material.

[0017] As a preferred solution of the gas cylinder management method of the present invention, wherein: sending the relevant characteristics into a graph convolutional neural network model for fusion includes:

[0018] Based on the graph convolutional neural network model, the first k basic acquisition features are regarded as graph nodes. The initial features of each graph node are generated through pre - coding. Meanwhile, the correlation between the features of each graph node is calculated, and the initial edge weights of the graph nodes are defined.

[0019] Update the graph node features and edge weights, and perform global average pooling on the features of the last updated graph node to obtain the fused features.

[0020] As a preferred solution of the gas cylinder management method described in the present invention, specifically: perform knowledge transfer according to the number of fused features to obtain the life prediction results of gas cylinders of each material, including:

[0021] Call the dataset D of different basic acquisition features under each material of the gas cylinder. If the number of fused features is less than the dataset D, knowledge transfer is adopted. If the number of fused features is equal to or greater than the dataset D, knowledge transfer is not adopted;

[0022] Based on the meta - learning model, train the deep residual network in the meta - learning model through the fused features to complete knowledge transfer;

[0023] Take out any two gas cylinders of different materials in the dataset D, and the data volumes of the two gas cylinders are in a proportional relationship, that is, B = A×k, where k is a non - zero constant, A represents the gas cylinder with scarce data volume, and B represents the gas cylinder with rich data volume;

[0024] Regard B as the source domain, and the source domain includes samples, training sets, and validation sets; regard A as the target domain, and the target domain includes samples, support sets, and query sets;

[0025] Divide the source domain into multiple training tasks, and each training task has different task difficulties;

[0026] Initialize the global parameters and local parameters of the deep residual network, perform local updates and global updates on each training task respectively, and update the global parameters and local parameters through the target domain.

[0027] As a preferred solution of the gas cylinder management method described in the present invention, specifically: updating the global parameters and local parameters through the target domain includes:

[0028] Perform local parameter updates on the support set of the target domain. Take the local parameter update process as the inner loop, evaluate the local parameter update by calculating the meta - loss on the query set of the target domain, take the process of calculating the meta - loss as the outer loop, and update the global parameters;

[0029] Repeat the inner loop and outer loop processes until the meta - loss reaches the convergence state, and output the life prediction results of gas cylinders of each material.

[0030] As a preferred solution of the gas cylinder management method described in the present invention, wherein: evaluating and optimizing the knowledge transfer effect, including:

[0031] In the case of adopting knowledge transfer, input the validation set into the meta-learning model to obtain the source domain prediction value. By calculating the error index MES, compare the source domain prediction value with the validation set to obtain the source domain accuracy; input the query set into the meta-learning model to obtain the target domain prediction value. By calculating the error index MES, compare the target domain prediction value with the query set to obtain the target domain accuracy;

[0032] Calculate the source domain accuracy and the target domain accuracy with the meta-learning model without knowledge transfer. If the calculation result is greater than 1, it means that knowledge transfer can improve the accuracy of the meta-learning model for predicting the life of gas cylinders of each material. Otherwise, it means that the knowledge transfer effect is not ideal and a deep residual network needs to be re-established.

[0033] In a second aspect, the present invention provides a gas cylinder management system, which includes:

[0034] A gas cylinder data processing module, configured to obtain time series data, material characteristics, and maintenance records from gas cylinders of each material. Based on the time series data, material characteristics, and maintenance records, use the PC algorithm to establish a gas cylinder life prediction task, and screen out relevant characteristics affecting the gas cylinder life according to the established gas cylinder life prediction task;

[0035] A gas cylinder life prediction module, configured to send the relevant characteristics into a graph convolutional neural network model for fusion, and perform knowledge transfer according to the number of fused characteristics to obtain the life prediction results of gas cylinders of each material;

[0036] A gas cylinder life prediction optimization module, configured to evaluate and optimize the knowledge transfer effect, so as to improve the accuracy of predicting the life of gas cylinders of each material and realize the intelligent management of gas cylinders.

[0037] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and wherein: when the processor executes the computer program, any step of the above method is implemented.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and wherein: when the computer program is executed by a processor, any step of the above method is implemented.

[0039] Compared with the prior art, the beneficial effects of the invention are:

[0040] 1. In the existing technology, the complex interaction between material properties and usage environment is often ignored in the prediction of gas cylinder life, resulting in limited generalization ability of the model. In contrast, the present invention establishes a gas cylinder life prediction task by adopting the PC (Peter-Clark) algorithm, screens out the key features affecting the gas cylinder life, constructs the dependence relationship between features through causal inference using the PC algorithm, combines time series data, material features, and maintenance records, accurately identifies the most important variables for gas cylinder life prediction, and improves the adaptability of the model to different types of gas cylinders.

[0041] 2. For rare material gas cylinders with scarce data volume, traditional transfer learning methods lack a cross-domain transfer mechanism for deep knowledge, resulting in insufficient robustness and accuracy of the model in small sample scenarios. The present invention designs a knowledge transfer scheme based on the quantity of fused features, and adopts a meta-learning model and a deep residual network (ResNet) to achieve deep knowledge transfer from gas cylinders with rich data to rare material gas cylinders. Among them, meta-learning optimizes global parameters through multi-task training and performs rapid adaptation on small samples in the target domain, while the deep residual network captures complex non-linear feature patterns. In this way, it helps to improve the accuracy and stability of rare material gas cylinder life prediction, filling the gap in the existing technology in small sample scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0043] Figure 1 is the overall flowchart of the gas cylinder management method according to an embodiment of the present invention;

[0044] Figure 2 is the influence diagram of feature selection on prediction MSE according to an embodiment of the present invention;

[0045] Figure 3 is the comparison diagram of prediction errors under different training sample sizes according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention with reference to the drawings of the specification. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways than those specifically described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0048] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0049] The present invention is described in detail in conjunction with schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally in a non-general proportion, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0050] At the same time, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner, and outer" is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0051] Unless otherwise clearly defined and limited in the present invention, the terms "installed, connected, and connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0052] Embodiment 1

[0053] Referring to Figure 1 , this is the first embodiment of the present invention, which provides a gas cylinder management method, including:

[0054] S1. Obtain time-series data, material characteristics, and maintenance records from gas cylinders of each material. Based on the time-series data, material characteristics, and maintenance records, use the PC algorithm to establish a gas cylinder life prediction task, and screen out relevant characteristics affecting the gas cylinder life according to the established gas cylinder life prediction task;

[0055] Specifically, timing data is obtained through a gas cylinder sensor and normalized.

[0056] Specifically, the timing data is directly read using the communication protocol (such as RS-485, Modbus, CAN bus, 4-20mA analog signal, etc.) supporting the sensor. If the gas cylinder sensor has been connected to a SCADA (Supervisory Control and Data Acquisition) system or an IoT platform (such as Alibaba Cloud IoT, AWS IoT, ThingsBoard, etc.), the historical data can be directly exported from the background data management of the system or platform.

[0057] It should be noted that the Z-score is used to normalize the timing data to avoid the problem that the numerical range differences of the dimensions in the timing data may cause the subsequent model learning to be biased towards large-scale features and thus ignore small-scale features.

[0058] Specifically, if the gas cylinder belongs to non-public facilities, the material characteristics of the gas cylinder are queried according to the material specifications in the gas cylinder nameplate or product manual (such as the material description of ISO 9809 standard gas cylinders). If the gas cylinder belongs to public facilities, the material characteristics of the gas cylinder are queried based on the registration records of the industry supervision platform.

[0059] Specifically, check the maintenance labels on the gas cylinder (such as the last inspection date, filling record, etc.) or according to the "Regulations on the Safety Supervision of Gas Cylinders", and the corresponding inspection agency will provide an electronic report on the gas cylinder safety monitoring, from which the gas cylinder maintenance record can be obtained.

[0060] It should be noted that the basic acquisition parameters include sampling frequency (such as once per second), data type (such as pressure, temperature, vibration, etc.), gas cylinder material characteristics (steel, aluminum, and composite materials), and gas cylinder maintenance records (maintenance cycle, maintenance time, etc.), and the timing data is included in the data type; for example, for a steel gas cylinder, pressure and temperature are collected by default; for a composite material gas cylinder, vibration data may need to be additionally collected; then for gas cylinders of different materials, the acquisition of their basic sampling parameters is increased and deleted according to their specific materials.

[0061] Furthermore, by defining different basic acquisition characteristics for gas cylinders of each material, a dataset D of different basic acquisition characteristics for gas cylinders of each material is formed.

[0062] Furthermore, through the PC algorithm, the causal relationship between the basic acquisition characteristics of gas cylinders of each material and their corresponding gas cylinder lifetimes is inferred from the dataset D, and a directed acyclic graph is generated; for example, if the directed acyclic graph consists of "pressure → temperature → lifetime", then temperature is the mediating variable.

[0063] It should be noted that the PC (Peter-Clark) algorithm is a classical algorithm for causal discovery, aiming to infer the causal structure between variables from observational data;

[0064] Furthermore, for each basic acquisition feature, calculate its average causal effect value and mutual information value corresponding to the life of the gas cylinder;

[0065] Specifically, calculate the average causal effect of each basic acquisition feature on the life of the gas cylinder, and the formula is expressed as:

[0066] ACE(X i →Y) = E[Y∣do(X i = x1)] - E[Y∣do(X i = x0)]

[0067] Where, X i is a certain basic acquisition feature, x0 and x1 both represent the reference values of X i ; do represents the intervention operation; ACE represents the expected difference of a certain basic acquisition feature X i on the life of the gas cylinder Y;

[0068] Exemplarily, specify the relationship between x0 and x1 and X i . If X i is the pressure of the gas cylinder, then x0 can be the low-pressure state, such as 50 MPa (megapascal), and x1 can be the high-pressure state, such as 150 MPa (megapascal); if X i is the temperature of the gas cylinder, then x0 can be the normal temperature, such as 20 °C, and x1 can be the high temperature, such as 80 °C; this shows that x0 and x1 both represent the reference values of X i , and are two reference values with opposite states but representativeness;

[0069] Specifically, calculate the mutual information value between each basic acquisition feature and the life of the gas cylinder, and the formula is expressed as:

[0070]

[0071] Where, MI represents the mutual information value between a certain basic acquisition feature X i and the life of the gas cylinder Y. The larger the value of MI(X i , Y), the stronger the dependence between the basic acquisition feature and the life of the gas cylinder. If MI(X i , Y) = 0, it means that the two are independent; p(x i , y) represents the joint probability that X i takes the value of x i and Y takes the value of y, and x i represents X iAll the values of, including x0 and x1, and similarly y represents all the values of Y; for example, the probability that the pressure is 50 MPa and the gas cylinder life is 10 years.

[0072] Furthermore, by the average causal effect value and the mutual information value, an importance score is defined, and the top k basic acquisition features with the highest importance scores are selected.

[0073] Specifically, the importance score S i is expressed as:

[0074] S i = w1·ACE i + w2·MI i

[0075] where w1 and w2 correspond to the weight values of ACE i and MI i respectively;

[0076] Specifically, k is determined by the total number of basic acquisition features and the number of gas cylinders of each material.

[0077] It should be noted that by "linking" k (i.e., the top k basic acquisition features selected) with the total number of basic acquisition features and the number of gas cylinders of each material, the selection of basic acquisition features can be dynamically adjusted according to the data volume of gas cylinders of different materials, thus realizing the "personalized" configuration of the selection of basic acquisition features; because in practical applications, the data volumes of gas cylinders of different materials are often unbalanced. For example, there are fewer samples of some rare materials (such as carbon fiber composite gas cylinders), while there are more samples of common materials (such as steel gas cylinders). Then, for steel gas cylinders with rich data volume, more features (a larger k value) can naturally be selected.

[0078] S2. Send the relevant features into the graph convolutional neural network model for fusion, and perform knowledge transfer according to the number of fused features to obtain the life prediction results for gas cylinders of each material.

[0079] Furthermore, based on the graph convolutional neural network model, the top k basic acquisition features are regarded as graph nodes, the initial features of each graph node are generated by pre-encoding, and at the same time, the correlation between the features of each graph node is calculated to define the initial edge weights of the graph nodes.

[0080] Specifically, the cosine similarity is used to calculate the correlation between the features of each graph node, and the initial edge weights of the graph nodes are obtained as:

[0081]

[0082] where, represents the initial edge weights, Denote the initial feature of the n-th graph node generated by pre-coding, Denote the initial feature of the m-th graph node generated by pre-coding;

[0083] Furthermore, update the graph node features and edge weights, perform global average pooling on the features of the last updated graph node to obtain the fused feature;

[0084] Specifically, the update of the graph node features is expressed by the formula:

[0085]

[0086] where N(n) is the neighbor nodes of the graph node, d n is the degree of the n-th graph node, d m is the degree of the m-th graph node, W (l) Denote as the weight matrix, σ = ReLU is the activation function; l and l + 1 respectively represent the l-th layer and the (l + 1)-th layer of the graph convolutional neural network;

[0087] Specifically, the update of the edge weights is expressed by the formula:

[0088]

[0089] where || denotes the concatenation operation, W e is the training parameter;

[0090] Specifically, the fused feature H is expressed by the formula:

[0091]

[0092] where L represents the total number of layers of the graph convolutional neural network, N represents the total number of graph nodes;

[0093] Further, call the dataset D of different basic acquisition features under each material gas cylinder. If the number of fused features is less than the dataset D, knowledge transfer is adopted. If the number of fused features is equal to or greater than the dataset D, knowledge transfer is not adopted;

[0094] It should be noted that assume the called dataset D has 15 features and the number of fused features is only 8 features. If these 8 features are highly correlated with the gas cylinder life prediction (such as pressure trend, fatigue data), then knowledge transfer may still be effective; conversely, if the number of fused features reaches 20 but contains many irrelevant features (such as the influence of environmental humidity on steel gas cylinders is small), then the transfer may not work well; then on this basis, first screen the basic acquisition features through importance scores, then the correlation problem of the fused features can be well solved, thus making the transfer effective;

[0095] Furthermore, based on the meta-learning model, by fusing features, the deep residual network in the meta-learning model is trained to complete knowledge transfer;

[0096] Specifically, the deep residual network contains multiple residual blocks, and the form of each block is I = F(H) + H. F(H) represents the residual mapping, which is composed of multiple convolutional layers, activation functions, etc. H is the input fused feature, and I is the output fused feature;

[0097] Furthermore, any two gas cylinders of different materials are taken out from the dataset D, and the data volumes of the two gas cylinders of different materials are in a proportional relationship, that is, B = A × k, where k is a non-zero constant, A represents the gas cylinder with a small data volume, and B represents the gas cylinder with a large data volume;

[0098] Furthermore, B is regarded as the source domain, and the source domain includes samples, training sets, and validation sets; A is regarded as the target domain, and the target domain includes samples, support sets, and query sets;

[0099] It should be explained that the target domain is a specific domain or task that the model can adapt to and make predictions in knowledge transfer, and the source domain is usually a rich dataset; taking the prediction of the gas cylinder life as an example, currently, we want to predict the life of a carbon fiber composite gas cylinder, but the relevant data samples are scarce. Compared with the large data volume of steel gas cylinders (source domain), then we need to apply the knowledge in the source domain to the target domain through transfer learning. The support set and the query set are equivalent to "textbooks" and "exam questions". Suppose the support set has 10 data samples of carbon fiber composite gas cylinders, and each sample includes features (pressure, temperature, etc.) and the corresponding life (such as 25,000 hours); then through the support set, the model can learn the characteristics of these 10 data samples of carbon fiber composite gas cylinders, that is, form a "textbook", and then through the query set, that is, the "exam questions", evaluate the results predicted by the model after learning the samples, that is, check whether the "exam questions" are as guided by the "textbook";

[0100] Furthermore, the source domain is divided into multiple training tasks, and each training task has a different task difficulty;

[0101] Furthermore, the global parameter θ and the local parameter θ′ of the deep residual network are initialized, and local updates and global updates are performed on each training task respectively, and the global parameters and local parameters are updated through the target domain;

[0102] It should be noted that updating the global parameters and local parameters of each training task through the target domain is a key step to make the model adapt to the target domain;

[0103] Furthermore, local parameter updates are performed on the support set of the target domain. The local parameter update process is used as the inner loop. By calculating the meta-loss on the query set of the target domain, the local parameter updates are evaluated. The process of calculating the meta-loss is used as the outer loop, and the global parameters are updated.

[0104] Specifically, before the update, not only the global parameters and local parameters of the deep residual network need to be initialized, but also the global learning rate and local learning rate need to be initialized. The global learning rate is set to 0.001, and the local learning rate is set to 0.01.

[0105] Specifically, local parameter updates are performed on the support set of the target domain, which can be expressed by the formula:

[0106]

[0107] where α represents the local learning rate, α = {α0, α1, …, α n}; represents the loss value of the nth task under the support set support, represents the gradient of the global parameters, θ′ n represents the nth local parameter update value;

[0108] Specifically, the meta-loss L is calculated on the query set of the target domain meta which is expressed by the formula:

[0109]

[0110] where represents the loss value of the nth task under the query set query;

[0111] Specifically, the update of the global parameters is expressed by the formula:

[0112]

[0113] where η represents the global learning rate;

[0114] Specifically, the inner loop and outer loop processes are repeated until the meta-loss reaches the convergence state, and the life prediction results of gas cylinders of each material are output.

[0115] Furthermore, the local learning rate is adjusted according to the task difficulty. Define the variance of the training task as D n , then we get:

[0116] α n = α0·exp(D n ), α0 = 0.01

[0117] where α0 is the initialized local learning rate;

[0118] It should be noted that by adjusting the local learning rate according to the task difficulty, the model reduces the learning rate in tasks with a large number of features to avoid overfitting, while maintaining a high learning rate in tasks with constant or small numbers of features to accelerate the convergence of the meta-loss;

[0119] S3. By evaluating and optimizing the knowledge transfer effect, the accuracy of the life prediction of gas cylinders of each material is improved, and the intelligent management of gas cylinders is realized;

[0120] Furthermore, in the case of using knowledge transfer, the validation set is input into the meta-learning model to obtain the source domain prediction value. By calculating the error index MES, the source domain prediction value is compared with the validation set to obtain the source domain accuracy; the query set is input into the meta-learning model to obtain the target domain prediction value. By calculating the error index MES, the target domain prediction value is compared with the query set to obtain the target domain accuracy;

[0121] Specifically, the formula for the error index MES is as follows:

[0122]

[0123] Among them, v represents the number of the validation set or the query set, represents the source domain prediction value or the target domain prediction value, z i represents the validation set or the query set; the output result of the error index MES is the source domain accuracy or the target domain accuracy;

[0124] Even further, the source domain accuracy and the target domain accuracy are calculated with the meta-learning model without using knowledge transfer. If the calculation result is greater than 1, it indicates that knowledge transfer can improve the accuracy of the meta-learning model for the life prediction of gas cylinders of each material. Otherwise, it indicates that the knowledge transfer effect is not ideal and the deep residual network needs to be re-established;

[0125] Specifically, the source domain accuracy and the target domain accuracy are calculated with the meta-learning model without using knowledge transfer, and the formula is as follows:

[0126]

[0127] Among them, TLE represents the transfer coefficient, MSE source represents the source domain accuracy, MSE base represents the meta-learning model without using knowledge transfer, MSE target represents the target domain accuracy;

[0128] It should be noted that the accuracy of the meta-learning model for predicting the life of gas cylinders of each material is improved through knowledge transfer. According to the predicted life of the gas cylinders, the life of gas cylinders of each material is uniformly managed respectively, which can not only reduce the workload of manual inspection, but also identify and replace aging or damaged gas cylinders, thereby reducing the risk of accident events caused by gas cylinder failures and providing effective support for intelligent gas cylinder management.

[0129] Furthermore, this embodiment also provides a gas cylinder management system, including:

[0130] A gas cylinder data processing module, configured to obtain time series data, material characteristics, and maintenance records from gas cylinders of each material, establish a gas cylinder life prediction task using the PC algorithm based on the time series data, material characteristics, and maintenance records, and screen out relevant characteristics affecting the gas cylinder life according to the established gas cylinder life prediction task;

[0131] A gas cylinder life prediction module, configured to fuse the relevant characteristics by sending them into a graph convolutional neural network model, and perform knowledge transfer according to the number of fused characteristics to obtain the life prediction results of gas cylinders of each material;

[0132] A gas cylinder life prediction optimization module, configured to evaluate and optimize the knowledge transfer effect, thereby improving the accuracy of predicting the life of gas cylinders of each material and realizing the intelligence of gas cylinder management.

[0133] This embodiment also provides a computer device applicable to the gas cylinder management method, including:

[0134] A memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the gas cylinder management method proposed in the above embodiment.

[0135] This computer device can be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0136] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the gas cylinder management method proposed in the above embodiment.

[0137] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0138] Embodiment 2

[0139] Refer to Figure 2 and Figure 3 , which is the second embodiment of the present invention. This embodiment provides a gas cylinder management method, including:

[0140] In this embodiment, three gas cylinders made of different materials are selected as test objects: a steel gas cylinder (AISI 4130), an aluminum gas cylinder (AA 6061-T6), and a carbon fiber composite gas cylinder (Toray T800H / 3900-2); the test data is from the gas cylinder monitoring system of an industrial gas company, covering the usage records from 2022 to 2025. The data acquisition devices include Honeywell TruStability pressure sensors (model: HSC series, accuracy ±0.1% FS), Omega temperature sensors (model: THW-4, accuracy ±0.5°C), and a maintenance record database (based on MySQL 8.0); the control group uses a traditional linear regression model to construct a multiple linear regression equation to predict the lifespan based on historical pressure, temperature data, and maintenance cycles. By using an SVM (support vector machine) model, the data of the steel gas cylinder is used as the source domain and transferred to the target domain (aluminum and carbon fiber gas cylinders) through feature matching.

[0141] The experimental group adopted the solution of the present invention, and the steps are as follows: Z-score standardization was used to eliminate the dimensional differences in the time-series data. The material characteristics of the gas cylinders were obtained through a material handbook, including the yield strength (steel: 550 MPa, aluminum: 275 MPa, carbon fiber: 2,800 MPa) and the corrosion resistance grade (ISO 9223 standard); they were encoded as binary features according to the maintenance cycle (6 months / 12 months) and the test results (qualified / unqualified); the PC algorithm was used to construct a causal relationship graph, and the importance scores were calculated (weights: ACE 60%, MI 40%), and the top 5 key features were selected (for steel cylinders: pressure fluctuation, temperature gradient, filling times, maintenance cycle, fatigue index; for carbon fiber cylinders: vibration amplitude, temperature change rate, ultraviolet exposure duration, resin thickness, filling pressure); the graph nodes were initialized, and each feature node was embedded as a 128-dimensional vector. The edge weights were calculated through cosine similarity. The network structure: 2-layer graph convolution (GCN), the activation function was ReLU, and the pooling layer output a 256-dimensional fusion feature; the source domain (steel gas cylinders, with 10,000 data records) and the target domain (carbon fiber gas cylinders, with 200 data records) were divided according to the ratio B = 50×A, and the task difficulty adjusted the local learning rate (α 0 = 0.01); the training parameters: the global learning rate η = 0.001, iterated 50 times, and the meta-loss convergence threshold was 0.01;

[0142] By comparing the experimental group and the control group, some data were presented in the form of experimental graphs. Referring to Figure 2 , it can be seen that as the number of features decreased from the complete feature set to the selected number, the MSE of the three types of material gas cylinders (steel, aluminum, carbon fiber) all decreased significantly. When the number of features of the steel gas cylinders increased from 5 to 10, the MSE decreased from a higher value to a lower value, and similar trends were also shown for the aluminum and carbon fiber gas cylinders; it shows that the PC algorithm removed the noise features by selecting the top k features with the highest importance scores, improving the prediction accuracy of the model. The results were consistent with the effect after the PC algorithm selected features, proving the advantage of the present invention in feature selection; secondly, referring to Figure 3 , in the small sample scenario (10, 20 samples) and the medium sample scenario (50, 100 samples), due to insufficient data, the prediction error of the non-transfer model was large and the performance was poor; while the transfer model utilized the knowledge of the source domain and significantly reduced the error, reflecting the advantage of knowledge transfer;

[0143] In summary, through experimental comparison, it was further proved the strong applicability of the solution of the present invention in the environment of scarce data.

[0144] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0145] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0146] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0147] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0148] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0149] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to cover these changes and modifications.

Claims

1. A method for gas cylinder management, characterized in that, Including: Obtain time series data, material characteristics, and maintenance records from gas cylinders of each material. Based on the time series data, material characteristics, and maintenance records, use the PC algorithm to establish a gas cylinder life prediction task. According to the established gas cylinder life prediction task, screen out relevant characteristics that affect the gas cylinder life. Send the relevant characteristics into a graph convolutional neural network model for fusion. According to the number of fused characteristics, perform knowledge transfer to obtain the life prediction results for gas cylinders of each material. Evaluate and optimize the knowledge transfer effect, thereby improving the accuracy of life prediction for gas cylinders of each material and realizing the intelligent management of gas cylinders.

2. The gas cylinder management method according to claim 1, characterized in that, Using the PC algorithm, establish a gas cylinder life prediction task. According to the established gas cylinder life prediction task, screen out relevant characteristics that affect the gas cylinder life, including: Define different basic acquisition characteristics for gas cylinders of each material to form a dataset D of different basic acquisition characteristics for gas cylinders of each material. Through the PC algorithm, infer the causal relationship between the basic acquisition characteristics and the corresponding gas cylinder life for gas cylinders of each material from the dataset D, and generate a directed acyclic graph. For each basic acquisition characteristic, calculate its average causal effect value and mutual information value corresponding to the gas cylinder life. Define an importance score through the average causal effect value and the mutual information value, and select the top k basic acquisition characteristics with the highest importance score. Among them, k is determined by the total number of basic acquisition characteristics and the number of gas cylinders of each material.

3. The gas cylinder management method according to claim 2, characterized in that, Send the relevant characteristics into a graph convolutional neural network model for fusion, including: Based on the graph convolutional neural network model, regard the top k basic acquisition characteristics as graph nodes. The initial characteristics of each graph node are generated through pre-coding, and at the same time, calculate the correlation between the characteristics of each graph node, and define the initial edge weights of the graph nodes. Update the graph node characteristics and edge weights, and perform global average pooling on the characteristics of the last updated graph node to obtain the fused characteristics.

4. The gas cylinder management method according to claim 2 or 3, characterized in that, According to the number of fused characteristics, perform knowledge transfer to obtain the life prediction results for gas cylinders of each material, including: Call the dataset D of different basic acquisition characteristics for gas cylinders of each material. If the number of fused characteristics is less than the dataset D, use knowledge transfer. If the number of fused characteristics is equal to or greater than the dataset D, do not use knowledge transfer. Based on the meta-learning model, through the fused characteristics, train the deep residual network in the meta-learning model to complete knowledge transfer. Take out any two gas cylinders of different materials from the dataset D, and the data volumes of the two gas cylinders of different materials are in a proportional relationship, that is, B = A × k, where k is a non-zero constant, A represents the gas cylinder with scarce data volume, and B represents the gas cylinder with rich data volume. Regard B as the source domain, and the source domain includes samples, training sets, and validation sets; regard A as the target domain, and the target domain includes samples, support sets, and query sets. Divide the source domain into multiple training tasks, and each training task has different task difficulties. Initialize the global parameters and local parameters of the deep residual network, perform local updates and global updates on each training task respectively, and update the global parameters and local parameters through the target domain.

5. The gas cylinder management method according to claim 4, characterized in that Updating the global parameters and local parameters through the target domain, including: Performing local parameter update on the support set of the target domain, taking the local parameter update process as the inner loop, evaluating the local parameter update by calculating the meta-loss on the query set of the target domain, taking the process of calculating the meta-loss as the outer loop, and updating the global parameters; Repeating the inner loop and outer loop processes until the meta-loss reaches the convergence state, and outputting the life prediction results of gas cylinders of each material.

6. The gas cylinder management method according to claim 4, wherein Evaluating and optimizing the knowledge transfer effect, including: In the case of adopting knowledge transfer, inputting the validation set into the meta-learning model to obtain the source domain prediction value, calculating the error index MES, comparing the source domain prediction value with the validation set to obtain the source domain accuracy; inputting the query set into the meta-learning model to obtain the target domain prediction value, calculating the error index MES, and comparing the target domain prediction value with the query set to obtain the target domain accuracy; Calculating the source domain accuracy and the target domain accuracy with the meta-learning model without knowledge transfer. If the calculation result is greater than 1, it indicates that knowledge transfer can improve the accuracy of the meta-learning model for predicting the life of gas cylinders of each material. Otherwise, it indicates that the knowledge transfer effect is not ideal and a deep residual network needs to be re-established.

7. A gas cylinder management system, based on the gas cylinder management method according to any one of claims 1 to 6, characterized in that, Including: A gas cylinder data processing module configured to obtain time series data, material characteristics, and maintenance records from gas cylinders of each material, establish a gas cylinder life prediction task based on the time series data, material characteristics, and maintenance records, and adopt the PC algorithm, and screen out relevant features affecting the life of gas cylinders according to the established gas cylinder life prediction task; A gas cylinder life prediction module configured to fuse the relevant features by sending them into a graph convolutional neural network model, and perform knowledge transfer according to the number of fused features to obtain the life prediction results of gas cylinders of each material; A gas cylinder life prediction optimization module configured to evaluate and optimize the knowledge transfer effect, thereby improving the accuracy of predicting the life of gas cylinders of each material and realizing the intelligent management of gas cylinders.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 6.

Images