Equipment state evaluation method, model training method thereof and electronic equipment

Abstract

The invention provides a state evaluation method of equipment, a model training method of the equipment and electronic equipment. The network architecture of the state evaluation model comprises a bidirectional gating circulation unit network and an attention mechanism layer; the training method comprises the following steps: acquiring a time sequence sample, wherein the time sequence sample comprises multi-frame equipment operation data marked with an equipment marking state; inputting the time sequence sample into a state evaluation model, and carrying out additive fusion on multi-dimensional parameters contained in the equipment operation data by an attention mechanism layer; inputting the additive fusion result into a two-way gating circulation unit network, extracting equipment state characteristics by the two-way gating circulation unit network, and outputting an equipment prediction state; and according to the loss error of the equipment labeling state and the equipment prediction state, adjusting model parameters of the state evaluation model. The state evaluation model is formed by combining the attention mechanism layer and the two-way gating circulation unit network, and the overall precision and reliability of the state evaluation model can be greatly improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of device status assessment, and in particular to a device status assessment method and a model training method thereof, and an electronic device. Background Art

[0002] During the use of equipment, the performance may gradually degrade due to wear, aging or other factors. When the performance degrades to a certain extent, the equipment may fail. When the equipment begins to degrade or fail, there is a lag in the change of equipment energy consumption, that is, the change in energy consumption will not be immediately apparent. This means that the increase or change in energy consumption may not become obvious until the equipment performance has degraded for a period of time. This lag means that the performance change of the equipment may not be discovered in time through the current energy consumption data alone, and the performance degradation of the equipment is uncertain, making it difficult to predict the future during the equipment status assessment process. Summary of the invention

[0003] The technical problem to be solved by the present disclosure is to overcome the defect of poor accuracy of device status evaluation in the prior art, and to provide a device status evaluation method and a model training method thereof, and an electronic device.

[0004] The present invention solves the above technical problems through the following technical solutions:

[0005] In a first aspect, a training method for a state assessment model is provided, wherein the network architecture of the state assessment model includes a bidirectional gated recurrent unit network and an attention mechanism layer; the training method includes:

[0006] Acquire a time series sample, wherein the time series sample includes multiple frames of device operation data, and each frame of device operation data is annotated with a device annotation state;

[0007] Inputting the time series samples into the state assessment model, so that the attention mechanism layer performs additive fusion on the multi-dimensional parameters contained in the device operation data to obtain an additive fusion result;

[0008] Inputting the additive fusion result into a bidirectional gated recurrent unit network, so that the bidirectional gated recurrent unit network extracts device state features, and outputs a device predicted state according to the device state features;

[0009] According to the loss error determined by the device annotation state and the device prediction state, the model parameters of the state assessment model are adjusted until an iteration stop condition is met.

[0010] Optionally, obtain time series samples, including:

[0011] Determine the maximum value of the data sampling frequency of each dimensional parameter, and determine the target time window according to the maximum value;

[0012] Using the target time window to perform sliding window processing on the original sampling data of the device to obtain multiple frames of data segments;

[0013] The time series samples are constructed according to the multiple frame data segments.

[0014] Optionally, constructing the time series sample according to the multiple frame data segments includes:

[0015] For each dimension parameter contained in each frame data segment, other parameter values ​​except the mean value or the maximum value in the data segment are eliminated;

[0016] Determine the data segments that have undergone data elimination processing as the device operation data; or splice the data segments that have undergone data elimination processing to a preset time length, and determine the data segments of the preset time length as the device operation data;

[0017] The device marking state is determined according to the device state corresponding to the timestamp of each dimensional parameter retained in the device operation data.

[0018] Optionally, using the target time window to perform sliding window processing on the original sampling data of the device includes:

[0019] In response to the presence of missing values ​​in the original sampled data, the missing values ​​are interpolated and filled using the following formula:

[0020]

[0021] Among them, x t represents the filling result of missing values ​​at time t; x t-a represents the sampling data at time ta; x t+b Represents the sampling data at time t+b; T t+b Indicates the timestamp at time t+b; T t Indicates the timestamp at time t; T t-a Indicates the timestamp of the moment.

[0022] Optionally, the attention mechanism layer adopts a Bahdanauz attention mechanism layer.

[0023] In a second aspect, a method for evaluating a device status is provided, comprising:

[0024] Obtain the operation data sequence of the equipment;

[0025] Inputting the operation data sequence into a state assessment model to determine the device state of the time period corresponding to each frame of operation data contained in the operation data sequence according to the state assessment model; wherein the state assessment model is trained according to the training method of the state assessment model described in any one of the first aspects;

[0026] The health status of the device is evaluated according to the device status in each time period.

[0027] Optionally, the health status of the device is evaluated by the following formula:

[0028]

[0029] Among them, H represents health; T represents the length of the running data sequence; output k Indicates the device status in the time period corresponding to the k-th frame of operating data contained in the operating data sequence.

[0030] Optionally, it also includes:

[0031] Establishing a corresponding relationship between each time period and the device status in the time period;

[0032] The health status of the device in a future time period is predicted according to the corresponding relationship.

[0033] According to a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein the processor implements any one of the methods described in the first aspect when executing the computer program.

[0034] According to a fourth aspect, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of the first aspects is implemented.

[0035] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0036] The positive progress of the present disclosure is that the attention mechanism layer of the state evaluation model in the present disclosure automatically performs additive fusion on the multi-dimensional parameters contained in the device operation data, thereby preventing some parameter features from being discarded during the feature dimension iteration process in the device state recognition task due to their own low value (but a large contribution to the task after being combined with other parameter features); the bidirectional gated recurrent unit network of the state evaluation model has the ability to introduce the short-term future state of the device into machine learning, and to analyze and evaluate the device state in combination with the historical operation data and current operation data of the device. Therefore, the combination of the attention mechanism layer and the bidirectional gated recurrent unit network can greatly improve the overall accuracy and reliability of the state evaluation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of a training method for a state assessment model provided by an exemplary embodiment of the present disclosure;

[0038] Figure 2 A schematic diagram of the structure of an attention mechanism layer used in a training method of a state assessment model provided by an exemplary embodiment of the present disclosure;

[0039] Figure 3 A schematic diagram of the structure of a bidirectional gated recurrent unit network used in a training method of a state assessment model provided by an exemplary embodiment of the present disclosure;

[0040] Figure 4 A flow chart of a method for evaluating a state of a device provided by an exemplary embodiment of the present disclosure;

[0041] Figure 5 The present invention provides a schematic structural diagram of an electronic device according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0043] Prefixes such as "first" and "second" are used in the embodiments of the present disclosure only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers to distinguish description objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and no unnecessary limitation should be constituted due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0044] Figure 1 A flowchart of a training method for a state assessment model provided by an exemplary embodiment of the present disclosure, wherein the network architecture of the state assessment model includes a bidirectional gated recurrent unit network and an attention mechanism layer. Figure 1 , the training method includes:

[0045] Step 101: Obtain time series samples.

[0046] The time series sample includes multiple frames of equipment operation data, and each frame of equipment operation data is annotated with the equipment annotation status. The number of frames of equipment operation data included in the time series sample can be set according to actual conditions.

[0047] The representation of the equipment labeling status can be set according to the actual situation. For example, it can be expressed by abnormal energy consumption and normal energy consumption. The status assessment model is used to evaluate the energy consumption status of the equipment. The output parameters of the status assessment model include abnormal energy consumption and normal energy consumption. Or the quantitative value "1" indicates complete abnormality, "0" indicates complete health, and the intermediate value between 0 and 1 indicates abnormality. The output parameters of the status assessment model are quantitative values ​​in the range of [0,1].

[0048] Equipment operation data includes parameters in at least two dimensions: equipment energy consumption, equipment detection point data, equipment environment, output, working hours, equipment attributes, etc. Equipment detection point data includes detection point voltage, equipment detection point current, equipment detection point vibration value, etc. Equipment environment includes temperature, humidity, etc. Equipment attributes include rated power, energy efficiency, etc.

[0049] The time series samples constructed by parameters of multiple dimensions are used as training samples to train the state assessment model, so that the state assessment model has the ability to extract correlation features from parameters of multiple dimensions, and can conduct a comprehensive and integrated analysis of the equipment status, thereby improving the sensitivity of the state assessment model. And the state assessment model is trained with time series samples (sequence data), and the time series samples can be divided into three stages of data: historical equipment operation data, current equipment operation data, and short-term future equipment operation data, so that the state assessment model can learn the equipment status change trend based on the time series, and thus have the ability to reasonably predict the future status of the equipment.

[0050] Step 102: Input the time series samples into the state assessment model, so that the attention mechanism layer performs additive fusion on the multi-dimensional parameters contained in the equipment operation data to obtain an additive fusion result.

[0051] In one embodiment, see Figure 2 , the attention mechanism layer adopts the Bahdanauz attention mechanism layer. Taking the device operation data containing parameters h and s, and inputting them into the Bahdanauz attention mechanism layer as an example, the Bahdanauz attention mechanism layer first uses two parameter matrices W to input the parameters h and s. q (q*d) and W k (k*d) are mapped to d dimensions respectively, and then they are added (elementwise), after a tanh activation function, they are dot-producted with another parameter matrix W (also d dimensions), and finally the attention weight a (scalar) is obtained, which is then normalized using softmax. This splicing process can help the neural network quickly learn the combined features between multiple variables, thereby solving the problems of unstable coupling features, difficult feature screening, and affecting model accuracy in the process of equipment status assessment.

[0052] Step 103: Input the additive fusion result into the bidirectional gated recurrent unit network so that the bidirectional gated recurrent unit network extracts the device state features and outputs the device predicted state according to the device state features.

[0053] Understandably, the input dimension of the bidirectional gated recurrent unit network is determined by the output dimension of the attention mechanism layer.

[0054] In one embodiment, see Figure 3 The number of cells (n_cell) of the bidirectional gated recurrent unit network is set to n*t*5, the output layer is set to a fully connected layer, the activation function uses softmax, and the evaluation criterion is accuracy. The basic unit of the bidirectional gated recurrent unit network is the GRU unit. Each GRU unit contains an update gate and a reset gate to control the flow of information. The bidirectional gated recurrent unit model introduces a bidirectional recurrent structure and divides the GRU unit into two parts, one part processes the sequence data from front to back, and the other part processes the sequence data from back to front. In this way, the model can capture past and future contextual information at the same time, so as to better understand the sequence data. Specifically, the input of the bidirectional gated recurrent unit network is a sequence data. First, the input sequence data is converted into a vector representation of fixed dimension through the embedding layer. Then, the vector representation is input into the BiGRU layer. The BiGRU layer consists of two independent GRU layers, one processing the sequence data from front to back and the other processing the sequence data from back to front. The outputs of the two GRU layers are concatenated to form the final output.

[0055] Step 104: According to the loss error determined by the device annotation state and the device prediction state, the model parameters of the state assessment model are adjusted until the iteration stop condition is met.

[0056] The loss function in step 104 may be, but is not limited to: MAE mean absolute error function, MSE mean square error function, RMSE square root function, DTW function, etc.

[0057] In this embodiment, the parameters of the bidirectional gated recurrent unit network and the attention mechanism layer are iteratively optimized together. During the feature dimension iteration process, the features actually required in the use process are selected based on the accuracy and stability of the test results during the training process.

[0058] The iteration stop condition can be set according to the actual situation, including but not limited to: the number of iterations reaches the number threshold, the loss error is less than the error threshold. The number threshold and error threshold can be set according to the actual situation.

[0059] The trained status assessment model can identify the abnormal status of equipment at any time based on time series data.

[0060] In this embodiment, the attention mechanism layer of the state assessment model automatically performs additive fusion on the multi-dimensional parameters contained in the device operation data, thereby preventing some parameter features from being discarded during the feature dimension iteration process in the device state recognition task due to their own low value (but a large contribution to the task after being combined with other parameter features); the bidirectional gated recurrent unit network of the state assessment model has the ability to introduce the short-term future state of the device into machine learning, and to analyze and evaluate the device state in combination with the historical operation data and current operation data of the device. Therefore, the combination of the attention mechanism layer and the bidirectional gated recurrent unit network can greatly improve the overall accuracy and reliability of the state assessment model.

[0061] In one embodiment, a time window is dynamically determined according to the data sampling frequency of each dimensional parameter contained in the device operation data, and sliding window processing is performed on the original sampling data of the device based on the time window to obtain multiple frames of data segments, and a time series sample is constructed based on the multiple frames of data segments. Step 101 specifically includes:

[0062] Step 101 - 1 : determine the maximum value of the data sampling frequency of each dimensional parameter, and determine the target time window according to the maximum value.

[0063] Assume that the sampling frequencies of the dimensional parameters s1, s2, …sn are f1, f2, …fn respectively. Select the maximum value f* of the sampling frequency and construct a time window with a width of 2 / f*. This time window is the target time window. Where n represents the total number of dimensions of the parameters contained in the equipment operation data.

[0064] For example, the equipment operation data includes parameters of 16 dimensions (n=16), wherein the maximum value of the sampling frequency is 1 Hz, and the width of the target time window is 2 s.

[0065] It should be noted that according to the sampling law and industry knowledge and experience, when the sampling frequency is higher than twice the highest frequency of the parameter signal, the signal characteristics can be better preserved and restored. Of course, the width of the time window is not limited to twice 1 / f*, and the width of the time window can also be other multiples of 1 / f*, but too large a width of the time window will cause redundancy in computing power and storage space, and too small a width will cause information loss and affect the accuracy of subsequent models. Preferably, the width of the time window is twice 1 / f*.

[0066] Step 101 - 2: Use the target time window to perform sliding window processing on the original sampling data of the device to obtain multiple frames of data segments.

[0067] The timestamps of the parameter signals of each dimension are not synchronized. Before sliding window processing, the timestamps need to be aligned. In one implementation, the parameter signal with the maximum value f* of the sampling frequency is used as the alignment reference. If the timestamps of other parameter signals are different from the reference signal, the value closest to the timestamp is taken. For example, the maximum value f* of the sampling frequency uses the data of the parameter corresponding to f* as the time window reference. When sliding window processing is performed, the data of the parameters of each dimension contained in the original sampling data of the device is sliced ​​into m time windows with a width of 2 / f*. Each time window obtains a frame of data segment, and a total of m frames of non-overlapping data segments are obtained. Taking the width of the target time window as 2s as an example, the duration of each frame of data segment is 2s.

[0068] Step 102-3: construct a time series sample according to the multiple frame data segments.

[0069] In this embodiment, the timestamps of the various dimensions of parameters contained in the original sampling data of the device are out of sync. In the process of constructing the training samples of the state assessment model, the parameter data is divided and merged through a dynamic time window to ensure the synchronization of the timestamps of the various dimensions of parameters, thereby improving the robustness of the state assessment model.

[0070] In one embodiment, when constructing a time series sample, the data is also preprocessed, such as cleaning the original data. The main cleaning steps may include but are not limited to: a. filling in missing values; b. eliminating a small number of outliers through interquartile range statistical analysis; c. deleting redundant fixed value data.

[0071] The following is an implementation method for filling missing values: in response to the presence of missing values ​​in the original sampling data, the missing values ​​are interpolated and filled using the following formula:

[0072]

[0073] Among them, x t represents the filling result of missing values ​​at time t; x t-a represents the sampling data at time ta; x t+b Represents the sampling data at time t+b; T t+b Indicates the timestamp at time t+b; T t Indicates the timestamp at time t; T t-a Indicates the timestamp of the moment.

[0074] In one embodiment, if the data of a certain parameter is relatively stable data, the average value or factory value of the data of the parameter is determined as the filling result of the missing value, and interpolation filling is performed.

[0075] In one embodiment, the device labeling status of each frame of device operation data included in the time series sample in step 101 is manually labeled by an expert.

[0076] The following introduces another implementation method for automatically determining the annotation status of a device: For each dimension parameter included in each frame of data segment, eliminate other parameter values in the data segment except for the mean value or the maximum value, that is, retain the mean value or the maximum value, so as to perform dimensionality reduction processing on the data segment; determine the data segment after data elimination processing as the device operation data to obtain the sample set SA; determine the device annotation status according to the device status corresponding to the timestamps of each dimension parameter retained in the device operation data. For example, align the timestamps of the sample set SA and calculate the average value L as the device annotation status of the sample set SA.

[0077] The following introduces another implementation method for automatically determining the annotation status of a device: For each dimension parameter included in each frame of data segment, eliminate other parameter values in the data segment except for the mean value or the maximum value; splice the data segment after data elimination processing to a preset time length t, where t represents the time step of the time series sample input into the state evaluation model each time, which is equivalent to extending and reconstructing the data segment. The preset time length t can be set according to the actual situation, for example, (1 < t <= 10). Determine the data segment with the preset time length as the device operation data, and determine the device annotation status according to the device status corresponding to the timestamps of each dimension parameter retained in the device operation data.

[0078] Figure 4 The flowchart of a method for evaluating the status of a device provided by an exemplary embodiment of the present disclosure. The method for evaluating the status of the device includes the following steps:

[0079] Step 401, obtain the operation data sequence of the device.

[0080] The duration of the operation data sequence can be set according to the actual situation. For example, the duration is 100s, that is, collect 100s of device operation data to construct the operation data sequence for device status evaluation.

[0081] Step 402, input the operation data sequence into the state evaluation model to determine the device status corresponding to the time period of each frame of operation data included in the operation data sequence according to the state evaluation model.

[0082] Among them, the state evaluation model is trained according to the training method of the state evaluation model provided in any of the above embodiments. The state evaluation model can accurately identify the device status in each time period.

[0083] Step 403, evaluate the health status of the device according to the device status in each time period.

[0084] In one embodiment, the health status of the device is evaluated by the following formula:

[0085]

[0086] Wherein, H represents the health degree; T represents the duration of the operation data sequence; output k represents the device state corresponding to the time period of the k-th frame of operation data included in the operation data sequence. The duration of T can be set according to actual needs, for example, 5 < T < 100.

[0087] In this embodiment, the device state in different time periods is quantified according to the state evaluation model, and then the overall health state of the device is comprehensively analyzed based on the device states in different time periods, so as to improve the accuracy of device state evaluation.

[0088] In one embodiment, a health degree threshold F can be set (the default is 50%, which can be adjusted according to actual situations and requirements). If H < F, the device is in an abnormal state and requires manual maintenance, and an alarm prompt is issued; otherwise, the device is in a normal state and no action is taken.

[0089] In one embodiment, the state evaluation method further includes: constructing a correspondence between each time period and the device state in the time period, and predicting the health state of the device in the future time period according to the correspondence.

[0090] Wherein, the correspondence can be represented by a curve or by a function.

[0091] When the correspondence is represented by a curve, the device operation data collected in the previous t - 1 time of the current moment and the device operation data at the current moment t are used as the input of the state evaluation model, and the device states at different moments from 0 to t are obtained, fitted into a curve, and the trend of the curve is extended to predict the health state at the next moment.

[0092] Corresponding to the foregoing embodiments of the training method of the state evaluation model and the state evaluation method of the device, the present disclosure also provides embodiments of a training device for the state evaluation model and a state evaluation device for the device.

[0093] The embodiment of the present disclosure provides a training device for a state evaluation model. The network architecture of the state evaluation model includes a bidirectional gated recurrent unit network and an attention mechanism layer; the training device includes:

[0094] An acquisition module, configured to acquire a time series sample, where the time series sample includes multiple frames of device operation data, and each frame of device operation data is labeled with a device labeled state;

[0095] An input module, configured to input the time series sample into the state evaluation model, so that the attention mechanism layer performs additive fusion on multi-dimensional parameters included in the device operation data to obtain an additive fusion result;

[0096] The input module is further used to input the additive fusion result into a bidirectional gated recurrent unit network, so that the bidirectional gated recurrent unit network extracts device state features, and outputs a predicted device state according to the device state features;

[0097] The training module is used to adjust the model parameters of the state assessment model according to the loss error determined by the device annotation state and the device prediction state until the iteration stop condition is met.

[0098] Optionally, the acquisition module is specifically used for:

[0099] Determine the maximum value of the data sampling frequency of each dimensional parameter, and determine the target time window according to the maximum value;

[0100] Using the target time window to perform sliding window processing on the original sampling data of the device to obtain multiple frames of data segments;

[0101] The time series samples are constructed according to the multiple frame data segments.

[0102] Optionally, when constructing the time series samples according to the multiple frames of data segments, the acquisition module is specifically used to:

[0103] For each dimension parameter contained in each frame data segment, other parameter values ​​except the mean value or the maximum value in the data segment are eliminated;

[0104] Determine the data segments that have undergone data elimination processing as the device operation data; or splice the data segments that have undergone data elimination processing to a preset time length, and determine the data segments of the preset time length as the device operation data;

[0105] The device annotation state is determined according to the device state corresponding to the timestamp of each dimensional parameter retained in the device operation data.

[0106] Optionally, when the target time window is used to perform sliding window processing on the original sampling data of the device, the acquisition module is specifically used to:

[0107] In response to the presence of missing values ​​in the original sampled data, the missing values ​​are interpolated and filled using the following formula:

[0108]

[0109] Among them, x t represents the filling result of missing values ​​at time t; x t-a represents the sampling data at time ta; x t+b Represents the sampling data at time t+b; T t+b Indicates the timestamp at time t+b; T t Indicates the timestamp at time t; Tt-a Indicates the timestamp of the moment.

[0110] Optionally, the attention mechanism layer adopts a Bahdanauz attention mechanism layer.

[0111] The present disclosure provides a device for evaluating a state of a device, including:

[0112] An acquisition module, used to acquire the operation data sequence of the device;

[0113] An evaluation module, used for inputting the operation data sequence into a state evaluation model, so as to determine the device state of the time period corresponding to each frame of operation data contained in the operation data sequence according to the state evaluation model; wherein the state evaluation model is trained by a training device for a state evaluation model according to any one of claims 1 to 5;

[0114] The health status of the device is evaluated according to the device status in each time period.

[0115] Optionally, the health status of the device is evaluated by the following formula:

[0116]

[0117] Among them, H represents health; T represents the length of the running data sequence; output k Indicates the device status in the time period corresponding to the k-th frame of operating data contained in the operating data sequence.

[0118] Optionally, it also includes:

[0119] A construction module, used to construct a corresponding relationship between each time period and the device status in the time period;

[0120] The prediction module is used to predict the health status of the equipment in a future time period according to the corresponding relationship.

[0121] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution.

[0122] Figure 5This is a schematic diagram of the structure of an electronic device showing an example embodiment of the present disclosure, the electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor, and the processor implements the method described in any of the above embodiments when executing the computer program. Figure 5 The electronic device 50 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0123] like Figure 5 As shown, the electronic device 50 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 50 may include, but are not limited to: at least one processor 51, at least one memory 52, and a bus 53 connecting different system components (including the memory 52 and the processor 51).

[0124] The bus 53 includes a data bus, an address bus, and a control bus.

[0125] The memory 52 may include a volatile memory, such as a random access memory (RAM) 521 and / or a cache memory 522 , and may further include a read-only memory (ROM) 523 .

[0126] The memory 52 may also include a program tool 525 (or utility) having a set (at least one) of program modules 524, such program modules 524 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0127] The processor 51 executes various functional applications and data processing by running the computer program stored in the memory 52, such as the method provided in any of the above embodiments.

[0128] The electronic device 50 may also communicate with one or more external devices 54 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 55. Furthermore, the electronic device 50 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 56. As shown, the network adapter 56 communicates with other modules of the electronic device 50 via a bus 53. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 50, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0129] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0130] The embodiments of the present disclosure further provide a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the method provided in any of the above embodiments is implemented.

[0131] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0132] The present disclosure also provides a computer program product, including a computer program, which implements any of the above methods when executed by a processor.

[0133] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0134] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A training method for a state assessment model, characterized in that: The network architecture of the state assessment model includes a bidirectional gated recurrent unit network and an attention mechanism layer; the training method includes: Acquire a time series sample, wherein the time series sample includes multiple frames of device operation data, and each frame of device operation data is annotated with a device annotation state; Inputting the time series samples into the state assessment model, so that the attention mechanism layer performs additive fusion on the multi-dimensional parameters contained in the device operation data to obtain an additive fusion result; Inputting the additive fusion result into a bidirectional gated recurrent unit network, so that the bidirectional gated recurrent unit network extracts device state features, and outputs a device predicted state according to the device state features; According to the loss error determined by the device annotation state and the device prediction state, the model parameters of the state assessment model are adjusted until an iteration stop condition is met.

2. The training method of the state assessment model according to claim 1, characterized in that: Get time series samples, including: Determine the maximum value of the data sampling frequency of each dimensional parameter, and determine the target time window according to the maximum value; Using the target time window to perform sliding window processing on the original sampling data of the device to obtain multiple frames of data segments; The time series samples are constructed according to the multiple frame data segments.

3. The training method of the state assessment model according to claim 2, characterized in that: Constructing the time series sample according to the multiple frame data segments includes: For each dimension parameter contained in each frame data segment, other parameter values ​​except the mean value or the maximum value in the data segment are eliminated; Determine the data segments that have undergone data elimination processing as the device operation data; or splice the data segments that have undergone data elimination processing to a preset time length, and determine the data segments of the preset time length as the device operation data; The device annotation state is determined according to the device state corresponding to the timestamp of each dimensional parameter retained in the device operation data.

4. The training method of the state assessment model according to claim 2, characterized in that: The target time window is used to perform sliding window processing on the original sampling data of the device, including: In response to the presence of missing values ​​in the original sampled data, the missing values ​​are interpolated and filled using the following formula: Among them, x t represents the filling result of missing values ​​at time t; x t-a represents the sampling data at time ta; x t+b Represents the sampling data at time t+b; T t+b Indicates the timestamp at time t+b; T t Indicates the timestamp at time t; T t-a Indicates the timestamp of the moment.

5. The training method of the state assessment model according to any one of claims 1 to 4, characterized in that: The attention mechanism layer adopts the Bahdanauz attention mechanism layer.

6. A method for evaluating the state of an equipment, characterized in that: include: Obtain the operation data sequence of the equipment; Inputting the operation data sequence into a state assessment model to determine the device state of the time period corresponding to each frame of operation data contained in the operation data sequence according to the state assessment model; wherein the state assessment model is trained according to the training method of the state assessment model according to any one of claims 1 to 5; The health status of the device is evaluated according to the device status in each time period.

7. The device status assessment method according to claim 6, characterized in that: The health status of the device is evaluated by the following formula: Among them, H represents health; T represents the length of the running data sequence; output k Indicates the device status in the time period corresponding to the k-th frame of operating data contained in the operating data sequence.

8. The device status assessment method according to claim 6 or 7, characterized in that: Also includes: Establishing a corresponding relationship between each time period and the device status in the time period; The health status of the device in a future time period is predicted according to the corresponding relationship.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images