Predictive maintenance method and system for photovoltaic water heater based on deep learning

Abstract

The invention belongs to the technical field of photovoltaic water heater prediction, and provides a photovoltaic water heater predictive maintenance method and system based on deep learning in order to solve the problem of low generation power prediction accuracy, the method comprises the following steps: preprocessing historical information of a photovoltaic module and time sequence data of corresponding environment characteristics, and extracting multi-scale information; a multi-scale time sequence is decomposed into a seasonal sequence and a trend sequence by using a sequence decomposition module, and the seasonal sequence and the trend sequence are respectively mixed into multi-scale past information by using a past decomposable mixing module. And carrying out aggregation prediction by using a future multi-predictor hybrid module. According to the method, seasonal and trend components on different scales are decomposed and mixed to extract details and macroscopic trends in historical information, multi-scale information of micro seasons and macroscopic trends is fully utilized, and the accuracy of photovoltaic water heater photovoltaic panel power generation power prediction is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to photovoltaic water heater prediction, and in particular relates to a photovoltaic water heater predictive maintenance method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Solar energy is a clean, pollution-free renewable energy source that is widely distributed, inexhaustible and inexhaustible. Promoting the application of intelligent photovoltaic power generation operation and maintenance systems and using intelligent systems or technologies to replace manual operation and maintenance management can greatly reduce operation and maintenance costs and improve service efficiency.

[0004] The water source used in solar water heaters is mostly tap water, which is rich in minerals such as calcium and magnesium ions. During the heating process of water, the calcium and magnesium ions of the mineral species will react chemically with carbonate ions to form insoluble substances, which will be deposited on the surface of the heating tube of the water heater to form scale. The poor thermal conductivity of scale will hinder the transfer of heat, resulting in a decrease in heating efficiency, prolonged heating time, and increased energy consumption. In addition, long-term high-load operation of the heating device will accelerate the aging of solar water heater components and shorten their service life. The intelligent water heater operation and maintenance system can automatically determine the accumulation of scale and promptly remind users to clean the water heater, which is beneficial to human health and prolongs the service life of the equipment.

[0005] In addition, solar photovoltaic panels are mainly installed outdoors and exposed to sunlight for a long time. In actual use, the power generation efficiency of photovoltaic modules is not only affected by factors such as radiation, temperature, humidity and wind speed, but also easily affected by buildings, particle matter on the surface of modules, weeds, etc., resulting in low power generation efficiency of photovoltaic modules. The prediction of scale accumulation in water tanks and predictive maintenance of photovoltaic panels by traditional photovoltaic thermal solar energy is not mature yet, and mainly adopts manual inspection and regular cleaning. Therefore, it is impossible to achieve timely and accurate fault diagnosis and location.

[0006] There are currently few solutions to the problem of predictive maintenance of scale accumulation in the barrel of a solar water heater. In addition, machine learning and deep learning are now widely used, but since photovoltaic output is affected by many factors such as weather and light, the accuracy of the power generation prediction model is low. Summary of the invention

[0007] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a predictive maintenance method and system for photovoltaic water heaters based on deep learning, which decomposes and mixes seasonal and trend components at different scales to extract details and macro trends in historical information, fully utilizes the multi-scale information of micro-seasons and macro trends, and improves the accuracy of photovoltaic panel power generation prediction of photovoltaic water heaters.

[0008] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a photovoltaic water heater predictive maintenance method based on deep learning, comprising: Obtain the historical information of photovoltaic modules and the time series data of corresponding environmental characteristics, and perform preprocessing; The preprocessed time series data is downsampled to obtain a multi-scale time series; Use the sequence decomposition module to decompose the multi-scale time series into seasonal series and trend series; Using a past decomposable mixing module to mix the seasonal series and the trend series into multi-scale past information respectively; The future multi-predictor hybrid module is used to aggregate multi-scale past information and predict the power generation of photovoltaic panels of photovoltaic water heaters. Based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions, it is determined whether the photovoltaic water heater needs maintenance.

[0009] In a second aspect, the present invention provides a photovoltaic water heater predictive maintenance system based on deep learning, comprising: A preprocessing module is configured to: obtain historical information of photovoltaic components and time series data of corresponding environmental characteristics, and perform preprocessing; The scale module is configured to: downsample the preprocessed time series data to obtain a multi-scale time series; A decomposition module, configured to: decompose the multi-scale time series into a seasonal series and a trend series using a series decomposition module; A mixing module is configured to: use a past decomposable mixing module to mix the seasonal sequence and the trend sequence into multi-scale past information respectively; A prediction module is configured to: aggregate multi-scale past information using a future multi-predictor hybrid module to predict the power generated by the photovoltaic panel of the photovoltaic water heater; The maintenance module is configured to determine whether the photovoltaic water heater needs maintenance based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions.

[0010] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0011] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0012] One or more of the above technical solutions have the following beneficial effects: In the present invention, the time series data of the historical information of photovoltaic components and the corresponding environmental characteristics are preprocessed, and multi-scale information is extracted; the multi-scale time series is decomposed into seasonal series and trend series by using a sequence decomposition module, and the seasonal series and trend series are respectively mixed into multi-scale past information by using a past decomposable mixing module. Then, an aggregate prediction is performed using a future multi-predictor mixing module. The method of the present invention decomposes and mixes seasonal and trend components at different scales to extract details and macro trends in historical information, making full use of the multi-scale information of micro-seasons and macro trends, and improving the accuracy of photovoltaic panel power generation prediction of photovoltaic water heaters.

[0013] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0015] Figure 1 This is the schematic diagram of the existing solar electric water heater system; Figure 2 This is a flowchart of predictive maintenance of photovoltaic water heaters based on a deep learning model in Embodiment 1 of the present invention; Figure 3 This is a flow chart of data collection and organization in Embodiment 1 of the present invention; Figure 4 This is a flow chart of TimeMixer model prediction in Embodiment 1 of the present invention; Figure 5 This is a scale prediction alarm flow chart in the first embodiment of the present invention; Figure 6 This is a solar photovoltaic panel prediction alarm flow chart in the first embodiment of the present invention; Figure 7 This is a process diagram of selecting time series of different scales with higher weights in the first embodiment of the present invention.

[0016] Figure 8 (a)-(b) are respectively the network structure diagrams of seasonal mixing and trend mixing in the first embodiment of the present invention. DETAILED DESCRIPTION

[0017] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0018] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0019] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0020] Embodiment 1 This embodiment discloses a predictive maintenance method for photovoltaic water heaters based on deep learning, including: Obtain the historical information of photovoltaic modules and the time series data of corresponding environmental characteristics, and perform preprocessing; The preprocessed time series data is downsampled to obtain a multi-scale time series. Use the sequence decomposition module to decompose the multi-scale time series into seasonal series and trend series; Using a past decomposable mixing module to mix the seasonal series and the trend series into multi-scale past information respectively; The future multi-predictor hybrid module is used to aggregate multi-scale past information and predict the power generation of photovoltaic panels of photovoltaic water heaters. Based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions, it is determined whether the photovoltaic water heater needs maintenance.

[0021] This embodiment adopts a predictive maintenance method to plan a maintenance plan based on the predicted thermal efficiency of the electric heating and the photovoltaic power generation power of the photovoltaic water heater. It can effectively and timely treat scale, which is beneficial to human health and safety. At the same time, it eliminates the inconvenience caused by downtime for repairs when faults occur, and reduces the waste of manpower, material resources and financial resources caused by unnecessary and redundant regular maintenance. This embodiment achieves accurate prediction of scale accumulation in solar water heaters by calculating the useful work of the water heater when it is heated only by mains electricity and the total work used for heating by mains electricity for comparative analysis.

[0022] In order to predict the power generation efficiency of solar photovoltaic panels, a deep learning algorithm with powerful feature extraction and parallel computing capabilities is used to train a photovoltaic power generation data regression model based on time series prediction. The method proposed in this embodiment overcomes the problem of lack of predictive maintenance in traditional solar energy. By predicting the heat transfer efficiency of solar power heating and the power generation power of photovoltaic panels and comparing them with standard conditions, it can accurately locate the scale accumulation of solar water heaters and photovoltaic panel failures, and issue an early warning. Through this embodiment, the problem of slow heating of solar water heaters by power heating and low power generation efficiency of photovoltaic panels can be effectively solved.

[0023] This embodiment includes four steps: processing photovoltaic module data and weather forecast data, establishing a photovoltaic panel power generation prediction model, fault prediction diagnosis and tracking, and fault recovery determination. Among them, the first three steps are mainly the water heater fault prediction diagnosis process.

[0024] The following is a detailed description of a photovoltaic water heater predictive maintenance method based on deep learning proposed in this embodiment: Step 1: Data collection, text feature extraction and dimensionality reduction.

[0025] Step 101: Collect photovoltaic module data and weather forecast data as a time series data set.

[0026] The data set mainly includes weather, temperature, wind speed, wind direction, air humidity, irradiance, power, light intensity and other information. Among them, wind direction and weather features are semantic features, which need to be extracted from text features and converted into low-dimensional numerical feature vectors. Other features are all numerical features. The sample data can be regarded as multivariate time series data.

[0027] For semantic features, the torch framework in the PyCharm platform is used to import the pre-trained BERT model, encode the weather and wind direction features and output the text tensor, and import the PCA model for dimensionality reduction.

[0028] Specifically, BERT-Base and the Chinese library pre-training model are used. Each word is processed by the BERT model to obtain a 768-dimensional vector. The PCA algorithm is used for dimensionality reduction. Matrix transformation is used to convert multivariate time series data into feature-independent vectors, and features such as weather type and wind direction are converted into digital features.

[0029] Step 102: Use technologies such as the Internet of Things and WIFI to collect temperature and time variation data of the photovoltaic water heater when only the electric heating mode is turned on at night as a time series data set.

[0030] Step 2: Data processing.

[0031] Step 201: Use interpolation technology to interpolate missing time series values.

[0032] Make full use of the data points around the missing values ​​and use time-based interpolation, spline interpolation, linear interpolation, etc. to fill in the missing information.

[0033] Step 202: Perform denoising operation on the time series data.

[0034] The rolling mean method is used to calculate the mean for each ordered window, minimizing the noise in the time series data.

[0035] Step 203: Perform outlier detection on the data in the time series.

[0036] The unsupervised machine learning algorithm K-means is used to detect outliers in time series data. The distance from a data point to the nearest centroid is measured. If the distance is greater than a certain threshold, the data point is marked as an anomaly.

[0037] Step 204: Perform standard normalization on the data set:

[0038] in, is the feature mean, is the characteristic variance.

[0039] Step 205: Divide the data set into a training set, a validation set, and a test set. Randomly select 70% of the data as the training set, 20% of the data as the validation set, and the remaining 10% of the data for testing.

[0040] Step 3: Based on TimeMixer, the multi-scale mixed model can be decomposed to obtain the actual power generation of solar photovoltaic panels : Power generation under standard conditions It refers to the output power of power generation equipment under standard conditions. These standard conditions usually include specific environmental parameters such as temperature, pressure and humidity. It will vary according to the type and specification of the equipment. The power generation under standard conditions is It will be given in the factory manual.

[0041] Step 301: Pool past observations by average pooling Downsample to Different scales, finally obtain a set of multi-scale time series ,in, , Indicates the number of variables, the lowest level sequence is the input sequence containing the finest temporal variations, while the highest-level sequence For macro changes, K represents a specific number of data points.

[0042] Among them, the past observations This is the time series data processed in step 2.

[0043] Step 302: Project these multi-scale sequences to deep features through the embedding layer In the example, let it be the input of the first PDM module, which can be formalized as , and obtain a multi-scale representation of the input sequence.

[0044] Step 303: Using the past decomposable mixing module, namely the PDM module, the decomposed seasonal and trend components are mixed into multiple scales respectively.

[0045] Generally, for the layer, the input is , the PDM process can be formalized as:

[0046] in, Indicates the total number of layers, , Indicates that A mixture of channels is represented.

[0047] In this embodiment, the processing of the multi-scale sequence by the PDM module specifically includes: Step 303 - 1 : Use the routing function to select the top V most critical time series of different scales.

[0048] For the multi-scale time series that have been obtained , use a linear layer to transform the time series Map to the same length and form a matrix , use the routing function to get the weight of each mapped time series. According to the time series weight, select the first V time series of different scales with higher weights, and perform trend and season mixing. In the process of obtaining the time series weight, introduce the noise term to increase the randomness in the weight generation process. The specific operations are as follows:

[0049] in ， is the routing function, and are activation functions, and is the learning parameter used for weight generation, V>0, and is an integer.

[0050] According to the weight information obtained by the routing function, the first V time series with the largest weights of different scales are retained, that is, the (M+1) multi-scale time series obtained from the average pooling downsampling Select V time series of different scales ,like Figure 7 shown.

[0051] Step 303-2: Time series of layers Decomposition into seasonal time series and trend time series .

[0052] For multi-scale time series Decomposition into seasonal time series and trend time series Among them, The seasonal and trend decomposition in a PDM block can be formalized as:

[0053] in, For seasonal and trend decomposition modules, The most critical prerequisite for the routing function time series of different scales, == is an assignment operation.

[0054] Step 303-3: Using the -1) Layer multi-scale time series and The multi-scale time series obtained by mixing the seasonal term and trend term of the first layer is taken as Multiscale time series of layers:

[0055] in, It contains two linear layers, and the activation function GELU is included in the middle of the two linear layers for information interaction between channels. , denote seasonal mixing and trend mixing, respectively. Indicates The first PDM block The seasonal time series of scales, Indicates The first PDM block The trend term time series of scales, Indicates PDM blocks The time series of scale, Indicates the first ) time series of PDM blocks. Figure 8 (a)-(b) are schematic diagrams of seasonal mixing module and trend mixing module processing respectively.

[0056] Step 303-3: Incorporating information from low-level fine-scale time series into seasonal blending can provide additional details for coarse-scale seasonal modeling. , using a bottom-up blending layer in the The bottom-up interaction of seasonal information in a residual manner at each scale can be formalized as:

[0057] in, are two linear layers with an intermediate GELU activation function, whose input is seasonal time series with dimensions , the output is seasonal time series, the dimension becomes twice as large as before, that is, the output dimension is .

[0058] Step 303-4: In trend mixing, higher-level coarse-scale time series can more easily provide clear macro information. For the collection of multi-scale trend components , using a top-down mixing layer in the first The top-down trend information interaction is realized in a residual way at each scale, which can be formalized as:

[0059] in, are two linear layers with an intermediate GELU activation function, whose input is A trend time series with dimension , the output is seasonal time series, the dimension becomes half of the original, that is, the output dimension is .

[0060] Empowered by the hybrid of seasonality and trend, PDM gradually aggregates detailed seasonal information from fine to coarse, and fuses macro trend information with previous prior knowledge, ultimately achieving multi-scale hybridization in past information extraction.

[0061] 304: In After the PDM modules, multi-scale past information is obtained, which is expressed as , .

[0062] The Future Multi-Predictor Mixture (FMM) module is able to aggregate predictions from multi-scale sequences, making full use of multi-scale information, and can be expressed as:

[0063] in, Representatives from The future prediction in the scale sequence, the final output is . Refers to the The predictor of the scale sequence first uses multiple layers of stacked linearity and adds an activation function between each linear layer to directly regress the future of length F, from the length of The deep representation is then projected onto C variables. Based on different predictors from past information at different scales, the aggregate FMM is able to integrate the complementary prediction capabilities of mixed multi-scale sequences, and thus accurately predict the actual power generation of solar photovoltaic panels. .

[0064] Step 4: Compare the electric heating efficiency and photovoltaic power generation under normal conditions to determine whether there is a fault in the system. If there is a fault, push a fault alarm.

[0065] Step 401: comparing the thermal efficiency under the condition of only mains electricity heating in a normal state, and determining whether there is a fault in the electric heating system; Step 401-1: obtaining the time series data of the temperature and the corresponding time collected in step 1 under the condition that only the mains heating is turned on; Step 401-2: Calculate the useful power of the mains heating every day in 20 days:

[0066] in, , , They are the specific heat capacity of water, the density of water and the volume of the solar water heater tank. For the i Tiandi j The temperature when the mains heating is turned off for the first time, No. i Tiandi j The temperature when the mains heating is turned on for the first time ( ).

[0067] Step 401-3: Calculate the total power of the mains heating every day in 20 days. The specific calculation formula is as follows:

[0068] Where P is the rated power of the solar water heater. and Respectively i Tiandi j The time to turn off and on the AC heating.

[0069] Step 401-4: Calculate the average efficiency of 20-day mains heating , the specific calculation formula is as follows:

[0070] Step 401-5: Mains heating efficiency With threshold Compare, if it is lower than the preset threshold , it is determined that the mains heating system is faulty and a fault alarm is pushed to the user through the WeChat client; otherwise, the mains heating system is displayed as “normal”.

[0071] Step 402: comparing the photovoltaic power generation under normal conditions to determine whether there is a fault in the solar photovoltaic power generation system; Step 402-1: Use the prediction model obtained based on time series training in step 2 to predict the photovoltaic panel power generation function in the next 10 days according to weather conditions. , and obtain the actual power generation and time, perform polynomial fitting, and get the i Actual power generation per day as a function of time (i = 1,2,...,10); Step 402-2: Calculate the daily power generation required by the solar photovoltaic panels in the previous 10 days , Actual power generation ,

[0072]

[0073] in, , are the sunrise and sunset times of the i-th day, , They are respectively the power generation prediction function of the photovoltaic panel based on the weather conditions on the i-th day and the actual power generation fitting function based on the collected data; Step 402-3: Calculate the average power generation efficiency of the photovoltaic power generation system for 10 days , the specific calculation formula is as follows:

[0074] Step 402-4: Power generation efficiency With threshold If the value is lower than the preset threshold, the photovoltaic power generation system is judged to be faulty and a fault alarm is pushed to the user through the WeChat client. Otherwise, the photovoltaic power generation system is displayed as "normal". Step 5: Starting from the second day after the fault alarm is pushed, continue to re-determine the fault of the solar system: Step 501: electric heating system fault tracing; Step 501-1: Calculate the electric heating efficiency of the day and compare with the threshold; Step 501-2: If it is not lower than the preset threshold, the electric heating system fault alarm is cancelled. Otherwise, the fault alarm continues to be pushed to the user through the WeChat client.

[0075] Step 502: photovoltaic panel power generation system fault tracking; Step 502-1: Obtain the power generation of the day and the predicted power generation under standard conditions by querying data; Step 502-2: Calculate the power generation efficiency for the day and compare with the threshold; Step 502-3: If it is not lower than the preset threshold, the photovoltaic panel power generation system fault alarm is cancelled. Otherwise, the fault alarm continues to be pushed to the user through the WeChat client.

[0076] Embodiment 2 The purpose of this embodiment is to provide a photovoltaic water heater predictive maintenance system based on deep learning, including: A preprocessing module is configured to: obtain historical information of photovoltaic components and time series data of corresponding environmental characteristics, and perform preprocessing; The scale module is configured to downsample the preprocessed time series data to obtain a multi-scale time series. A decomposition module, configured to: decompose the multi-scale time series into a seasonal series and a trend series using a series decomposition module; A mixing module is configured to: use a past decomposable mixing module to mix the seasonal sequence and the trend sequence into multi-scale past information respectively; A prediction module is configured to: aggregate multi-scale past information using a future multi-predictor hybrid module to predict the power generated by the photovoltaic panel of the photovoltaic water heater; The maintenance module is configured to determine whether the photovoltaic water heater needs maintenance based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions.

[0077] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0078] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0079] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0080] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0081] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0082] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0083] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0084] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0085] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0086] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0087] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A predictive maintenance method for photovoltaic water heaters based on deep learning, characterized in that: include: Obtain the historical information of photovoltaic modules and the time series data of corresponding environmental characteristics, and perform preprocessing; The preprocessed time series data is downsampled to obtain a multi-scale time series; Use the sequence decomposition module to decompose the multi-scale time series into seasonal series and trend series; Using a past decomposable mixing module to mix the seasonal series and the trend series into multi-scale past information respectively; The future multi-predictor hybrid module is used to aggregate multi-scale past information and predict the power generation of photovoltaic panels of photovoltaic water heaters. Based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions, it is determined whether the photovoltaic water heater needs maintenance.

2. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: After downsampling the preprocessed time series data to obtain a multi-scale time series, the method further includes: using a linear layer to map the multi-scale time series to a matrix of the same length, using a routing function to obtain the weight of each mapped time series, and selecting the top V time series of different scales with the highest weight ranking according to the time series weight to perform trend and seasonal mixing; wherein V>0 and is an integer.

3. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: The seasonal series and the trend series are mixed into multi-scale past information using a past decomposable mixing module, specifically: The bottom-up mixing layer is used to achieve bottom-up interaction of seasonal series in a residual manner; A top-down mixing layer is used to realize the top-down trend sequence interaction in a residual manner.

4. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: It also includes the judgment of mains heating failure, specifically: Calculate the useful work and total work of mains heating; According to the calculated useful work and total work of the mains heating, the average efficiency of the mains heating is obtained; According to the comparison between the calculated average efficiency and the first set threshold value, it is determined whether the mains heating fails.

5. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: Based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions, the results of determining whether the photovoltaic water heater should be maintained are as follows: According to the predicted photovoltaic panel power generation power of the photovoltaic water heater, the photovoltaic panel power generation power function is fitted; According to the obtained photovoltaic panel power generation function, the expected power generation and actual power generation of the photovoltaic panel of the photovoltaic water heater are calculated; According to the expected power generation and actual power generation of the photovoltaic panels of the photovoltaic water heater, the average power generation efficiency of the photovoltaic panels of the photovoltaic water heater is calculated; The average power generation efficiency of the photovoltaic panel of the photovoltaic water heater is compared with the second set threshold value to determine whether the photovoltaic panel of the photovoltaic water heater is faulty.

6. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: The preprocessing includes: interpolation operation, denoising operation, outlier detection and normalization processing on the time series.

7. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: The future multi-predictor hybrid module uses stacked multiple linear layers to process past information of different scales, aggregates the processing results, and obtains the prediction result of the photovoltaic panel power generation of the photovoltaic water heater.

8. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1, characterized in that: The environmental characteristics include weather, temperature, wind speed, wind direction, air humidity, irradiance, power and light intensity.

9. A photovoltaic water heater predictive maintenance method based on deep learning as claimed in claim 1 or 8, characterized in that: It also includes semantic features in environmental features, extracting text features through the trained BERT model, and converting the extracted text features into numerical feature vectors.

10. A photovoltaic water heater predictive maintenance system based on deep learning, characterized in that: include: A preprocessing module is configured to: obtain historical information of photovoltaic components and time series data of corresponding environmental characteristics, and perform preprocessing; The scale module is configured to downsample the preprocessed time series data to obtain a multi-scale time series. A decomposition module, configured to: decompose the multi-scale time series into a seasonal series and a trend series using a series decomposition module; A mixing module is configured to: use a past decomposable mixing module to mix the seasonal sequence and the trend sequence into multi-scale past information respectively; A prediction module is configured to: aggregate multi-scale past information using a future multi-predictor hybrid module to predict the power generated by the photovoltaic panel of the photovoltaic water heater; The maintenance module is configured to determine whether the photovoltaic water heater needs maintenance based on the predicted photovoltaic panel power generation of the photovoltaic water heater and the electric heating efficiency of the photovoltaic water heater under normal conditions.

Images