Equipment prediction maintenance method, system and device based on time sequence algorithm

Abstract

The invention discloses an equipment prediction maintenance method based on a time sequence algorithm, and the method comprises the following steps: obtaining target data of electromechanical equipment, the target data comprising real-time data of a plurality of continuous time points; the real-time data comprises working condition data and operation data; inputting the target data into the prediction model to obtain predicted power data of a plurality of continuous time points in the future, and judging whether the electromechanical equipment needs to be maintained or not based on the predicted power data; wherein the prediction model is obtained by training based on a neural network model; the method comprises the steps of obtaining target data in electromechanical equipment, importing the target data into a prediction model to obtain predicted power data, judging whether scenic spot equipment needs to be maintained or not according to the predicted power data, and if it is judged that the electromechanical equipment needs to be maintained according to the predicted power data, arranging maintenance of the electromechanical equipment.

Description

Technical Field

[0001] The present invention relates to the field of building electromechanical operation and maintenance, and in particular to a method, system and device for predictive maintenance of equipment based on a timing algorithm. Background Art

[0002] In the era of deep integration of digitalization and intelligence, the management of mechanical and electrical equipment in buildings is undergoing unprecedented changes. With the rapid development of technologies such as the Internet of Things, big data, and cloud computing, the intelligence level of buildings is constantly improving, and higher requirements are placed on the AI ​​intelligent operation of mechanical and electrical equipment. However, the operation and maintenance methods of traditional building mechanical and electrical equipment still face many challenges. Although the traditional BA system has achieved a certain degree of automated control, its functions are single and its response is slow, making it difficult to adapt to the increasingly complex needs of modern building mechanical and electrical systems. In addition, the traditional operation and maintenance model relies too much on manual experience and preset rules, and lacks data-driven accurate decision-making capabilities, resulting in low operation and maintenance efficiency, untimely fault prediction, and serious energy waste. The mechanical and electrical systems in modern buildings, such as heating, ventilation and air conditioning (HVAC), elevators, water supply and drainage, etc., are not only large-scale and interrelated, but also have huge and complex operating data. How to efficiently use these data and mine the operating rules and fault warning information behind them has become the key to improving the operation and maintenance management level of building mechanical and electrical equipment. Summary of the invention

[0003] In order to overcome the shortcomings of the traditional operation and maintenance model of existing technology that relies too much on manual experience and preset rules, lacks data-driven accurate decision-making capabilities, resulting in low operation and maintenance efficiency, untimely fault prediction, and serious energy waste.

[0004] First aspect

[0005] The present invention provides a method for predictive maintenance of equipment based on a time series algorithm, comprising the following steps:

[0006] Acquire target data of the electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data;

[0007] The target data is input into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and whether the electromechanical equipment needs to be maintained is determined based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

[0008] Optionally, the training process of the prediction model includes:

[0009] A training sample set is obtained; the training sample set includes a plurality of training samples; the training samples include working condition data and operation data and corresponding equipment operation power;

[0010] The initialized neural network model is trained based on the training samples to obtain the prediction model.

[0011] Optionally, the training samples are preprocessed, and the preprocessing includes the following steps:

[0012] Sample cleaning: perform data cleaning on historical data to obtain primary data samples;

[0013] Sample screening: further screening the primary data samples according to the equipment business rules to obtain secondary data samples;

[0014] Sample elimination, removes outliers in the secondary data samples to obtain the training sample set.

[0015] Optionally, the actual power data of multiple consecutive time points in the future are monitored synchronously and in real time, a deviation reference value is set, and the deviation value between the actual power data and the predicted power data is calculated. When the deviation value exceeds the deviation reference value, a predictive maintenance result is generated.

[0016] Optionally, a classification threshold is set based on the deviation baseline value, and the predictive maintenance results are prioritized according to the classification threshold.

[0017] Optionally, based on the deviation reference value, a continuous exceeding time threshold is set, and if the deviation between the actual power data and the predicted power data continuously exceeds the deviation reference value and reaches the continuous exceeding time threshold, a maintenance warning is generated.

[0018] Optionally, predictive maintenance results are indicated by triggering an alarm or generating a work order.

[0019] Second aspect

[0020] The present invention provides a system, comprising:

[0021] The first module acquires target data of the electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data;

[0022] The second module is used to output the training sample set to the initial model, train the initial model based on the training sample set, and build a prediction model;

[0023] The third module is used to input the target data into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and determine whether the electromechanical equipment needs to be maintained based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

[0024] The third aspect

[0025] The present invention provides a device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement the steps of the method described in the first aspect when executing the program stored in the memory.

[0026] The beneficial effects of the present invention are: obtaining target data in electromechanical equipment, importing the target data into a prediction model to obtain predicted power data, judging whether scenic spot equipment needs maintenance through the predicted power data, and arranging maintenance for the electromechanical equipment if it is judged that the electromechanical equipment needs maintenance through the predicted power data. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0028] Figure 1 is a flowchart of a method for predictive maintenance of equipment based on a time series algorithm in some embodiments;

[0029] Figure 2 is another flow chart of a method for predictive maintenance of equipment based on a time series algorithm in some embodiments. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the concept, specific structure and technical effects of the present invention in combination with the embodiments and drawings, so as to fully understand the purpose, characteristics and effects of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by technicians in this field without creative work are all within the scope of protection of the present invention. In addition, all the connection / connection relationships involved in the patent do not refer to the direct connection of components, but refer to the formation of a better connection structure by adding or reducing connection accessories according to the specific implementation situation. The various technical features in the invention can be combined interchangeably without conflicting with each other.

[0031] like Figure 1 , 2 As shown, the present invention provides a device predictive maintenance method based on a time series algorithm, which is used to predict the operation and maintenance timing of electromechanical (such as a chiller) in a building, comprising the following steps:

[0032] S1. Acquire target data of electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data;

[0033] S2. Input the target data into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and determine whether the electromechanical equipment needs to be maintained based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

[0034] Execute step S1 to obtain target data of the electromechanical equipment. The target data can be obtained through transmission from a host computer or from a storage device, including real-time data at multiple consecutive time points. The real-time data includes historical operating data and operating data of the selected electromechanical equipment. The operating data includes but is not limited to data such as the outdoor temperature, passenger flow, and time type of the environment in which the electromechanical equipment is located. The operating data includes but is not limited to data such as the load rate, temperature, and vibration of the selected electromechanical equipment.

[0035] Execute step S2, input the target data into the prediction model, obtain the predicted power data for multiple consecutive time points in the future, and determine whether the electromechanical equipment needs to be maintained based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

[0036] The present application obtains target data from electromechanical equipment, imports the target data into a prediction model to obtain predicted power data, and determines whether scenic spot equipment needs maintenance based on the predicted power data. If the predicted power data determines that the electromechanical equipment needs maintenance, maintenance of the electromechanical equipment is arranged.

[0037] Furthermore, the target data may be real-time data at six consecutive time points of 10 minutes each in one hour, and the predicted power data at six time points of 10 minutes each in the next hour may be obtained.

[0038] In some embodiments, determining whether maintenance of electromechanical equipment is required based on the predicted power data includes:

[0039] S3. Synchronously monitor the actual power data of multiple consecutive time points in the future in real time, set the deviation reference value, calculate the deviation value between the actual power data and the predicted power data, and generate predictive maintenance results when the deviation value exceeds the deviation reference value.

[0040] Furthermore, real-time power data is monitored at 6 consecutive time points of 10 minutes each in the next hour.

[0041] S4. Based on the deviation reference value, a continuous exceeding time threshold is set. If the deviation between the actual power data and the predicted power data continuously exceeds the deviation reference value and reaches the continuous exceeding time threshold, a maintenance warning is generated.

[0042] Furthermore, the neural network model adopts an LSTM model, and the prediction model is obtained by obtaining basic data of electromechanical equipment within one year and importing it into the LSTM model to obtain an LSTM prediction model.

[0043] In some embodiments, the training process of the prediction model includes: obtaining a training sample set; the training sample set includes several training samples; the training samples include operating data and operation data as well as corresponding equipment operating power; and training the initialized neural network model based on the training samples to obtain the prediction model.

[0044] In some embodiments, the training samples are preprocessed, and the preprocessing includes the following steps:

[0045] Sample cleaning: perform data cleaning on historical data to obtain primary data samples;

[0046] Sample screening: further screening the primary data samples according to the equipment business rules to obtain secondary data samples;

[0047] Sample elimination: a multivariate linear regression model is used to train the device behavior pattern to generate a predicted value. By calculating the error between the actual value and the predicted value of the secondary data sample, samples whose errors exceed the predetermined proportion confidence interval are eliminated to obtain a normal training sample set.

[0048] During implementation, historical data is cleaned in sample cleaning, specifically, comprehensive historical data is cleaned, data missing, duplication, and card number problems are effectively handled, and over-limit, outlier, and abnormally turbulent data points are eliminated to obtain primary data samples, thereby ensuring data quality; in sample screening, primary data samples are further screened according to equipment business rules, specifically, data segments with turbulent equipment operation, initial startup, and significantly abnormal operating indicators are excluded to obtain secondary data samples to refine data input; in sample elimination, a multivariate linear regression model is used to train equipment behavior patterns to generate predicted values ​​for secondary data samples, and by calculating the error between actual values ​​and predicted values, samples whose errors exceed a predetermined proportion of the confidence interval are eliminated, where the predetermined proportion is 95%, and samples whose errors exceed the 95% confidence interval are eliminated, thereby generating a high-quality normal training sample set that meets the model training requirements, and through sample cleaning, data is screened based on business logic and a pre-processing solution based on a regression model to eliminate abnormal data, thereby ensuring the accuracy of the data used for subsequent analysis and the accuracy of the model.

[0049] In some embodiments, sample elimination can also be performed by setting an abnormal threshold percentage, identifying abnormal data in the secondary data sample that exceeds the abnormal threshold percentage and eliminating the abnormal data to obtain a training sample set.

[0050] During implementation, the random forest algorithm is used in combination with working conditions and operating data to set the abnormal threshold percentage, accurately identify and eliminate abnormal data, thereby generating a high-quality normal training sample set that meets the model training requirements. By setting the abnormal threshold, abnormal data can also be accurately identified and eliminated, thereby generating a high-quality training sample set.

[0051] In some embodiments, the process of training a neural network model includes the following steps:

[0052] Forward propagation, calculating the predicted value based on the time series data set;

[0053] Loss calculation, calculate the error between the predicted value and the actual value in the time series data set,

[0054] Back propagation, adjusting preliminary model parameters;

[0055] Parameter update, updating the weights and biases of the preliminary model;

[0056] Iterative training repeats the forward propagation, loss calculation, backpropagation and parameter update steps to generate a prediction model.

[0057] During implementation, the initial model uses the LSTM model. First, the model structure is set, including the number of LSTM layers, the number of units in each layer, etc.; confirm the prediction strategy, and clearly use the data of how many historical time points (such as the first 6 time points) as input to predict the data of how many time points in the future (such as the next 6 time points). Input the time series data set that meets the time interval determined by the prediction measurement in the preprocessing into the LSTM model. The training working principle is: 1. The input time series data set enters the LSTM model in chronological order; 2. At each time step, the LSTM model decides which old information to retain (forget gate), which new information to add (input gate), and what results to output (output gate); 3. These decisions are based on the current input and the previous hidden state. 4. Adjust the internal parameters of the LSTM, such as weights, to minimize the prediction error. Specific implementation steps (training process):

[0058] Forward propagation: input the training data into the LSTM network in time order; at each time step, the cell state and hidden state are calculated and updated through the LSTM unit; prediction is made using the final hidden state or the output of each time step (depending on the task requirements);

[0059] Loss calculation: Calculate the error between the predicted value and the actual value. Common loss functions include mean square error (MSE), cross entropy loss, etc.

[0060] Back propagation: Use the chain rule to calculate the gradient of the loss function with respect to the network parameters; since LSTM involves multiple time steps, it is necessary to update the gradient through time (BPTT).

[0061] Parameter update: The weights and biases of the LSTM network are updated using gradient descent (or its variants such as the Adam optimizer) to minimize the loss function.

[0062] Iterative training: Repeat the forward propagation, loss calculation, back propagation, and parameter update steps until the stopping condition is met (such as loss value convergence, reaching the maximum number of iterations, etc.).

[0063] The above steps finally lead to the LSTM time series prediction model. The LSTM (Long Short-Term Memory Network) time series prediction model is used to establish the time series relationship model between the equipment operation factors and the target parameters. Through the sliding window method, the LSTM model is trained with historical data within one year to confirm the input time point range and predict the target data at several time points in the future. The model can effectively capture the long-term dependency of equipment operation, improve the prediction accuracy, and provide a reliable basis for subsequent decision support.

[0064] In some embodiments, actual power data at multiple consecutive time points in the future are monitored synchronously and in real time, a deviation reference value is set, and the deviation value between the actual power data and the predicted data power is calculated. When the deviation value exceeds the deviation reference value, a predictive maintenance result is generated.

[0065] During implementation, the working conditions and operation data of electromechanical equipment at recent time points (such as the previous 6 time points) are collected and monitored in real time, such as the real-time data of 10 minutes every 1 hour, and brought into the LSTM time series prediction model trained in the previous step to predict the operation data of several time points in the future (such as the previous 6 time points), such as 10 minutes every 1 hour. And real-time monitoring of future operation data and predicted operation data, preset deviation benchmark values ​​(such as exceeding the upper limit and lower limit percentage thresholds), such as exceeding the upper limit by 10% is considered a large deviation, and health problems require predictive maintenance. By comparing the actual power data with the predicted power data to determine whether maintenance is needed, it can accurately predict the need for equipment maintenance, avoid excessive maintenance, and reduce the maintenance cost of the building.

[0066] Among them, the exceeding deviation reference value includes the upper percentage threshold of the exceeding deviation reference value and the lower percentage threshold of the exceeding deviation reference value.

[0067] If the prediction results show that the operating status of the equipment continues to deviate from the normal range and exceeds the preset threshold duration (such as exceeding the benchmark value every 10 minutes for 1 hour), the predictive maintenance recommendation will be automatically triggered, that is, an alarm will be triggered or a work order will be generated to indicate that the operating data exceeds the threshold, and the management personnel will be informed that the equipment has a potential failure risk and needs to be inspected and maintained in time.

[0068] In some embodiments, a continuous exceeding time threshold is set based on the deviation reference value, and if the deviation between the actual power data and the predicted power data continuously exceeds the deviation reference value and reaches the continuous exceeding time threshold, a maintenance warning is generated.

[0069] During implementation, by comparing the actual power and predicted power data, the deviation exceeds the preset threshold duration (such as exceeding the benchmark value every 10 minutes for 1 hour), and automatically triggering the predictive maintenance recommendation, that is, triggering an alarm or generating a work order, prompting the operation data to exceed the threshold, and informing the management personnel that the equipment has a potential failure risk and needs timely inspection and maintenance.

[0070] Furthermore, a grading threshold is set based on the deviation reference value, and the predictive maintenance results are prioritized according to the grading threshold. For example, if the deviation exceeds the reference value by 10%, it is defined as a minor abnormality and the cause of the casing needs to be checked. If the deviation exceeds the reference value by more than 50%, it is defined as a severe abnormality, requiring a maintenance warning and recommended repair. Severe and mild abnormalities are sorted, and severe abnormalities are given priority. The purpose of this is to effectively avoid trouble-free downtime maintenance. When it is defined as a minor abnormality, only the cause needs to be checked, and severe abnormalities require recommended repairs. At the same time, excessive maintenance is also avoided, and it is only necessary to judge whether maintenance is needed based on the system prompts.

[0071] At the same time, the system also supports the formulation and execution monitoring of maintenance plans, realizing intelligent operation and maintenance management. Through this method, operation and maintenance personnel can timely understand the status of equipment, take corresponding measures, reduce ineffective operation and maintenance and excessive maintenance, and improve equipment operation and maintenance efficiency.

[0072] The following is an example of a refrigeration station system in a building:

[0073] Obtain the hourly historical data of the refrigeration station system throughout the year. The historical data includes working condition data and operation data. The working condition data includes outdoor temperature, outdoor relative humidity, date type (working day, weekend), passenger flow, system cooling capacity, and the operation data system power data;

[0074] Data preprocessing includes: integrating the working condition data and operation data into a training sample set. First, the sample data is preprocessed for data quality (i.e., the sample cleaning mentioned above). When a parameter of the cooling station (one of the outdoor temperature, outdoor relative humidity, date type, human flow, system cooling capacity, and system power) is missing, the sample is deleted; when the values ​​of all the above parameters of the cooling station are the same, the first sample is retained; when the values ​​of three consecutive samples of the instantaneous data of the cooling station (one of the human flow, system cooling capacity, and system power) are the same, the first sample is retained; secondly, the data after data quality preprocessing is preprocessed for business logic (i.e., the sample screening mentioned above), and the samples when the system is shut down (system power) are removed. less than 5% of rated power), sample data 1 hour before and after the system switch is switched, and sample data of a parameter of the cold station that obviously exceeds the upper and lower limits (normal range of outdoor temperature -10~40, outdoor relative humidity 0-100, human flow 0-999999, system cooling capacity 0-999999, system power 0-999999); finally, the data after business logic preprocessing is initialized for modeling, and a multivariate linear regression model is used to build a regression model of outdoor temperature, outdoor relative humidity, date type (working day, weekend), human flow, system cooling capacity and system power, and samples with obvious deviations from the predicted value and the actual value are removed, such as samples with errors greater than the 95% quantile.

[0075] LSTM modeling is performed on the preprocessed samples. The first 6 hourly input data (outdoor temperature, outdoor relative humidity, date type, traffic volume, system cooling capacity, and system power) are selected to predict the output data (system power) of the next 6 hours. The LSTM standard time series model is used for modeling to obtain the trained time series prediction model.

[0076] Check whether the actual system power and the future predicted system power exceed a threshold value, such as 10%. If so, it is a minor abnormality and the cause can be investigated. If it exceeds 50%, a work order is generated to recommend maintenance.

[0077] The present invention provides a system, comprising:

[0078] The first module is used to obtain target data of electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data;

[0079] The second module is used to input the target data into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and determine whether the electromechanical equipment needs to be maintained based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

[0080] The present invention provides a device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement the steps of a device predictive maintenance method based on a timing algorithm when executing the program stored in the memory.

[0081] The present invention provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for predictive maintenance of equipment based on a time series algorithm are implemented.

[0082] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for predictive maintenance of equipment based on a time series algorithm, characterized in that: The following steps are involved: Acquire target data of the electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data; The target data is input into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and whether the electromechanical equipment needs to be maintained is determined based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

2. The equipment predictive maintenance method based on time series algorithm according to claim 1 is characterized in that: The training process of the prediction model includes: A training sample set is obtained; the training sample set includes a plurality of training samples; the training samples include working condition data and operation data and corresponding equipment operation power; The initialized neural network model is trained based on the training samples to obtain the prediction model.

3. The equipment predictive maintenance method based on time series algorithm according to claim 2 is characterized in that: The training samples are preprocessed, and the preprocessing includes the following steps: Sample cleaning: perform data cleaning on historical data to obtain primary data samples; Sample screening: further screening the primary data samples according to the equipment business rules to obtain secondary data samples; Sample elimination, removes outliers in the secondary data samples to obtain the training sample set.

4. The equipment predictive maintenance method based on time series algorithm according to claim 1 is characterized in that: The actual power data of multiple consecutive time points in the future are monitored synchronously and in real time, a deviation reference value is set, and the deviation value between the actual power data and the predicted power data is calculated. When the deviation value exceeds the deviation reference value, a predictive maintenance result is generated.

5. The equipment predictive maintenance method based on time series algorithm according to claim 4 is characterized in that: A classification threshold is set based on the deviation baseline value, and the predictive maintenance results are prioritized according to the classification threshold.

6. The equipment predictive maintenance method based on time series algorithm according to claim 4 is characterized in that: Based on the deviation reference value, a continuous exceeding time threshold is set. If the deviation between the actual power data and the predicted power data continuously exceeds the deviation reference value and reaches the continuous exceeding time threshold, a maintenance warning is generated.

7. The equipment predictive maintenance method based on time series algorithm according to claim 4 is characterized in that: Prompt predictive maintenance results by triggering alarms or generating work orders.

8. A system, characterized in that: include: The first module acquires target data of the electromechanical equipment, wherein the target data includes real-time data at multiple consecutive time points; the real-time data includes working condition data and operation data; In the second module, the target data is input into the prediction model to obtain the predicted power data for multiple consecutive time points in the future, and whether the electromechanical equipment needs to be maintained is determined based on the predicted power data; wherein the prediction model is obtained based on the training of the neural network model.

9. A device, characterized in that: The invention comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the steps of the method described in any one of claims 1 to 7 when executing the program stored in the memory.

Images