Remote intelligent operation and maintenance method and system based on photovoltaic wind power data

Abstract

The invention provides a remote intelligent operation and maintenance method and system based on photovoltaic wind power data, and the method comprises the steps: obtaining a plurality of original data sets of a plurality of wind and light stations, and carrying out the preprocessing of the original data sets, and forming a plurality of standby data sets; obtaining a time sequence length parameter and a space incidence matrix through the standby data set; constructing an initial output prediction model, updating the initial output prediction model into a final output prediction model through a time sequence length parameter and a space incidence matrix, and further obtaining historical prediction output data and real prediction output data; and acquiring prediction errors corresponding to the historical prediction output data, and expanding the real prediction output data into a real prediction output range through a plurality of prediction errors so as to judge whether a maintenance demand exists or not. Through the mode, end-to-end automatic remote intelligent operation and maintenance are realized, a means for monitoring the wind and light station in real time is provided, the consumption of manpower and material resources is reduced, the time cost is saved, and the real-time performance and accuracy of maintenance are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data prediction, in particular to a remote intelligent operation and maintenance method and system based on photovoltaic and wind power data. BACKGROUND

[0002] At present, energy power has gradually transformed from high-carbon, mainly fossil energy to low-carbon, mainly clean energy. With the continuous reduction of large-scale energy storage costs, wind and light stations have become the mainstream trend of new energy power due to their multi-time scale energy complementary characteristics.

[0003] The wind and light station uses a solar cell array and a wind turbine to store the generated electric energy in a storage battery pack. When the user needs to use electricity, the inverter converts the stored direct current in the storage battery pack into alternating current and sends it to the user load through the power transmission line.

[0004] The existing operation of the wind and light station is generally carried out in remote areas or unattended conditions. It is difficult and cumbersome to monitor and maintain the wind and light station distributed on the ground in real time, which requires a large amount of time cost and human and material resources, and it is difficult to ensure the real-time maintenance. SUMMARY

[0005] In view of the shortcomings of the prior art, the purpose of the present application is to provide a remote intelligent operation and maintenance method and system based on photovoltaic and wind power data, which aims to solve the technical problems that the wind and light station is located in a remote place and is distributed scattered, it is difficult and cumbersome to monitor and maintain it in real time, it requires a large amount of time cost and human and material resources, and it is difficult to ensure the real-time maintenance.

[0006] In order to achieve the above purpose, in a first aspect, the present application provides a remote intelligent operation and maintenance method based on photovoltaic and wind power data, comprising the following steps: Obtain a plurality of original data sets of a plurality of wind and light stations, the original data set comprising original output data and original meteorological data, and preprocess the original data set to obtain a standby data set comprising standby output data and standby meteorological data; Obtain a time series length parameter and a spatial correlation matrix corresponding to all wind and light stations through a plurality of standby output data; Construct an initial output prediction model, and construct an input data set and a result data set, update the initial output prediction model to a final output prediction model through the input data set, the result data set, the time series length parameter and the spatial correlation matrix, and obtain historical predicted output data through the final output prediction model; Obtaining a prediction error corresponding to the historical predicted output data, obtaining true predicted output data by the final output prediction model, expanding the true predicted output data into a true predicted output range by a plurality of the prediction errors, comparing true output data with the true predicted output range to determine whether there is a maintenance requirement.

[0007] Further, the step of preprocessing the original data set to obtain a to-be-used data set comprising to-be-used output data and to-be-used meteorological data comprises: Performing missing data filling and anomaly processing on the original output data and original meteorological data to obtain to-be-used output data and stage meteorological data; Performing normalization processing on the stage meteorological data to obtain to-be-used meteorological data, wherein the to-be-used output data and the to-be-used meteorological data constitute a to-be-used data set.

[0008] Still further, the to-be-used output data comprises to-be-used output values in a plurality of continuous time frames, and the step of obtaining a time series length parameter and a spatial correlation matrix corresponding to all wind-solar power stations by a plurality of the to-be-used output data comprises: Obtaining a station output mean value based on a plurality of the to-be-used output values, setting a plurality of delay orders, and obtaining a time series correlation coefficient corresponding to the delay order by the to-be-used output values, the station output mean value, and the delay order; Selecting a time series correlation coefficient greater than a first correlation threshold value as a candidate correlation coefficient, and selecting a delay order corresponding to the smallest candidate correlation coefficient as a time series length parameter; Obtaining a spatial correlation coefficient between two wind-solar power stations by the to-be-used output values and the station output mean value, and constructing a correlation relationship between the two wind-solar power stations with a spatial correlation coefficient greater than a second correlation threshold value to form a spatial correlation matrix.

[0009] Still further, the formula for obtaining the time series correlation coefficient is: , wherein, denotes the time series correlation coefficient, denotes a to-be-used output value in a tth time frame, denotes a station output mean value, denotes a delay order, denotes a total sampling duration; The formula for obtaining the spatial correlation coefficient is: , wherein, denotes a spatial correlation coefficient between an i th wind-solar power station and a j th wind-solar power station, represents the standby output value of the i th wind and light field station at the t th time frame, represents the standby output value of the j th wind and light field station at the t th time frame, represents the average station output of the i th wind and light field station, represents the average station output of the j th wind and light field station.

[0010] Further, the step of constructing the input data set and the result data set comprises: selecting a reference time node, and dividing the standby output data into a first data set and a result data set through the reference time node; dividing a second data set corresponding to the first data set from the standby meteorological data through the reference time node; splicing the first data set and the second data set into an input data set.

[0011] Further, the initial output prediction model comprises a spatial convolution module and a time attention module, and the step of updating the initial output prediction model to a final output prediction model through the input data set, the result data set, the time sequence length parameter and the spatial correlation matrix comprises: constructing a dynamic correlation matrix in the spatial convolution module through the input data set based on the spatial correlation matrix; converting the input data set into a stage feature set through the dynamic correlation matrix, the stage feature set comprising stage sub-features under a plurality of time sequences; setting a division window in the time attention module through the time sequence length parameter, inputting the stage feature set into the time attention module to select a to-be-updated sub-feature from a plurality of stage sub-features, and selecting a plurality of reference sub-features from a plurality of stage sub-features based on the division window, generating a final sub-feature corresponding to the to-be-updated sub-feature through the to-be-updated sub-feature and the plurality of reference sub-features; combining a plurality of final sub-features into a prediction output set, constructing a loss function through the prediction output set and the result data set to update the initial output prediction model to a final output prediction model.

[0012] Further, the step of constructing a dynamic correlation matrix in the spatial convolution module through the input data set comprises: obtaining an attention score between two wind and light field stations through the input data set; normalizing the attention score to obtain an attention weight between the two wind and light field stations; combining a plurality of attention weights into the dynamic correlation matrix.

[0013] Further, the attention score is obtained by the following formula: wherein, represents the attention score of the ith wind and light field station and the jth wind and light field station at the tth time frame, all represent learnable parameters, represents the transpose symbol, represents the time frame sub-data of the ith wind and light field station and the jth wind and light field station in the input data set at the tth time frame, represents an activation function.

[0014] Further, the step of expanding the real predicted output data into a real predicted output range by a plurality of prediction errors comprises: sorting a plurality of prediction errors, and setting a first position threshold and a second position threshold; obtaining the total number of a plurality of prediction errors, selecting a first error from a plurality of prediction errors by the total number and the first position threshold, and selecting a second error from a plurality of prediction errors by the total number and the second position threshold; superimposing the real predicted output data with the first error and the second error respectively to form a real predicted output range.

[0015] In a second aspect, the embodiments of the present application provide a remote intelligent operation and maintenance system based on photovoltaic and wind power data, which is applied to the remote intelligent operation and maintenance method based on photovoltaic and wind power data as described in the first aspect above, and the system comprises: a processing module configured to obtain a plurality of original data sets of a plurality of wind and light field stations, wherein the original data set comprises original output data, original weather data and original position data, and the original data set is preprocessed to obtain a standby data set comprising standby output data, standby weather data and standby position data; an extraction module configured to obtain a time series length parameter and a spatial correlation matrix corresponding to all wind and light field stations by a plurality of standby output data; a prediction module configured to construct an initial output prediction model, and construct an input data set and a result data set, update the initial output prediction model to a final output prediction model by the input data set, the result data set, the time series length parameter and the spatial correlation matrix, and obtain historical predicted output data by the final output prediction model; ​​​​​The execution module is configured to acquire a prediction error corresponding to the historical predicted output data, acquire real predicted output data by using the final output prediction model, expand the real predicted output data into a real predicted output range by using a plurality of prediction errors, compare real output data with the real predicted output range, and determine whether maintenance is required.

[0016] In a third aspect, an embodiment of the present application provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the remote intelligent operation and maintenance method based on photovoltaic and wind power data according to the first aspect when executing the computer program.

[0017] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, and the computer program is executable by a processor to implement the remote intelligent operation and maintenance method based on photovoltaic and wind power data according to the first aspect.

[0018] Compared with the prior art, the present application has the beneficial effects that: by acquiring the original output data and the original meteorological data between different wind and light field stations, constructing the time series length parameter and the spatial correlation matrix, the limitations of traditional independent monitoring of a single station are broken through, the correlation of output fluctuations between different wind and light field stations and the periodicity of output of the same wind and light field station are quantified, and the accuracy of output prediction is improved; by generating the real predicted output range based on the prediction error, the output uncertainty of the wind and light field station is quantified, a reasonable fluctuation range of output prediction is set, and the false alarm rate of maintenance is reduced; by constructing the final output prediction model, end-to-end automatic remote intelligent operation and maintenance is realized based on the foregoing method, a means for real-time monitoring of the wind and light field station is provided, the consumption of manpower and material resources is reduced, time cost is saved, and the real-time and accuracy of maintenance are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of the remote intelligent operation and maintenance method based on photovoltaic and wind power data in the first embodiment of the present application; Figure 2 A structural block diagram of the remote intelligent operation and maintenance system based on photovoltaic and wind power data in the second embodiment of the present application; The following specific embodiments will further illustrate the present application in combination with the above drawings. DETAILED DESCRIPTION

[0020] For the purpose of promoting an understanding of the application, the application will be described in greater detail for reference to the drawings. Several embodiments of the application are shown in the drawings. However, the application can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete. It is therefore contemplated that the application can be carried out in other ways than those specifically set forth herein without departing from the scope and spirit of the application.

[0021] It is noted that when a component is referred to as being "on" another component, it can be directly on the other component or intervening components can also be present. Where a component is referred to as being "connected" to another component, it can be directly connected to the other component or intervening components can also be present. The terms "vertical", "horizontal", "left", "right", and similar expressions as used herein are for illustrative purposes only.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0023] Referring to Figure 1 The first embodiment of the application provides a remote intelligent operation and maintenance method based on photovoltaic and wind power data, comprising the following steps: S10: obtaining a plurality of original data sets of a plurality of wind and light stations, the original data set comprising original output data and original meteorological data, and preprocessing the original data set to obtain a standby data set comprising standby output data and standby meteorological data; In this embodiment, the original output data is wind power or photovoltaic power. When the original output data is wind power, the original meteorological data includes wind speed and wind direction; when the original output data is photovoltaic power, the original meteorological data is irradiance and temperature. It can be understood that wind power and photovoltaic power can be processed respectively to complete the remote intelligent operation and maintenance of all wind and light stations. Further, the original output data comprises a plurality of sub-output data under continuous time frames, and the original meteorological data comprises a plurality of sub-meteorological data under continuous time frames, the sub-meteorological data corresponding to the sub-output data one by one.

[0024] The step S10 comprises: S110: performing missing data filling and abnormality processing on the original output data and original meteorological data to obtain standby output data and standby meteorological data; In the embodiment, the missing data is filled by using the average value of the adjacent time, and the abnormality processing refers to adaptive correction of the obviously wrong data, for example, if the sub-output data is less than 0, it is directly corrected to 0.

[0025] S120: normalizing the stage weather data to obtain weather data for use, the output data for use and the weather data for use constituting a data set for use; In the same weather data for use, the difference between the maximum value and the minimum value is taken as the denominator, and the difference between each value and the minimum value is taken as the numerator to complete the normalization processing. The normalization processing of the weather data can avoid the interference of the characteristic scale difference on the subsequent construction of the prediction model.

[0026] S20: obtaining a time series length parameter and a spatial correlation matrix corresponding to all wind and light stations through a plurality of output data for use; Specifically, the step S20 comprises: S210: obtaining a station output mean value based on a plurality of output values for use, setting a plurality of delay orders, and obtaining a time series correlation coefficient corresponding to the delay order through the output values for use, the station output mean value and the delay order; The formula for obtaining the time series correlation coefficient is: , Among them, denotes the time series correlation coefficient, denotes the output value for use under the t-th time frame, denotes the station output mean value, denotes the delay order, denotes the total sampling duration.

[0027] S220: selecting the time series correlation coefficient greater than the first correlation threshold value as a candidate correlation coefficient, and selecting the delay order corresponding to the smallest candidate correlation coefficient as the time series length parameter; In the embodiment, the first correlation threshold value is 0.6, and it should be noted that the delay order starts from 0 and increases, for example, a plurality of delay orders are 0, 1, 2, 3, 4, 5 and 6, and the time series correlation coefficients obtained are 0.1, 0.2, 0.4, 0.7, 0.8 and 0.9 respectively. At this time, the candidate correlation coefficients are 0.7, 0.8 and 0.9, and the time series length parameter corresponding to 0.7 is 4. Further, since multiple wind and light stations are processed synchronously, the time series length parameter corresponding to the largest number of multiple wind and light stations can be selected as the final value.

[0028] S230: obtain a spatial correlation coefficient between two wind-solar stations by the standby output value and the average station output value, and build a correlation relationship between two wind-solar stations with the spatial correlation coefficient greater than a second correlation threshold, to form a spatial correlation matrix; The formula for obtaining the spatial correlation coefficient is: , wherein, represents the spatial correlation coefficient between the ith wind-solar station and the jth wind-solar station, represents the standby output value of the ith wind-solar station at the tth time frame, represents the standby output value of the jth wind-solar station at the tth time frame, represents the average station output value of the ith wind-solar station, represents the average station output value of the jth wind-solar station. Assuming that there are 10 wind-solar stations, the 10 wind-solar stations can be selectively connected according to the spatial correlation coefficient.

[0029] S30: build an initial output prediction model, and build an input data set and a result data set, update the initial output prediction model to a final output prediction model by the input data set, the result data set, the time sequence length parameter and the spatial correlation matrix, and obtain historical prediction output data by the final output prediction model; The initial output prediction model includes a spatial convolution module and a time attention module.

[0030] The step 30 includes: S310: select a reference time node, and divide the standby output data into a first data set and a result data set by the reference time node; It can be understood that the standby output data includes standby output values at a plurality of continuous time frames. By taking the reference time node as a boundary, the standby output data can be divided into a plurality of standby output values in the early time and a plurality of standby output values in the late time. The plurality of standby output values in the early time form the first data set, and the plurality of standby output values in the late time form the result data set.

[0031] S320: cut out a second data set corresponding to the first data set from the standby meteorological data by the reference time node; It can be understood that the to-be-used meteorological data includes sub-meteorological data in a plurality of continuous time frames, and since the to-be-used output data corresponds to the time nodes in the to-be-used meteorological data one by one, after the to-be-used meteorological data is segmented, the sub-meteorological data earlier in time forms the second data set, and by introducing the second data set and synchronously inputting it into the prediction model with the first data set, it can be ensured that the to-be-used meteorological data that has a direct impact on the to-be-used output data is considered in the prediction process, and the prediction result is more accurate.

[0032] S330: splicing the first data set and the second data set into an input data set; S340: constructing a dynamic correlation matrix in the spatial convolution module based on the spatial correlation matrix and the input data set; By inputting the spatial correlation matrix, when the spatial convolution module updates the dynamic correlation matrix, only the data influence between wind and light stations with strong correlation needs to be considered, the complexity of calculation is reduced, the prediction model is avoided from being disturbed by unimportant noise relationships, and the prediction accuracy is improved. Specifically, the attention score between two wind and light stations is obtained through the input data set; The formula for obtaining the attention score is: , Among them, indicates the attention score of the ith wind and light station and the jth wind and light station at the tth time frame, 、 、 、 、 all indicate learnable parameters, indicates the transpose symbol, indicates the time frame sub-data of the ith wind and light station and the jth wind and light station in the input data set at the tth time frame, indicates an activation function. It should be noted that the learnable parameters can be iteratively updated by the loss function of the initial processing prediction model after setting the initial value, and then the final value is formed. Generally, the spatial convolution module has a plurality of convolution layers, so the attention score will be iterated through a plurality of convolution layers, is the output of the previous convolution layer, which is essentially the time frame sub-data of the ith wind and light station and the jth wind and light station in the input data set at the tth time frame. The time frame sub-data includes the to-be-used output value and the meteorological sub-data, representing synchronous processing of all types of input data. Through the spatial convolution module, the data of each wind and light station is fused with the information of other wind and light stations with strong correlation, improving the accuracy of data prediction.

[0033] The attention score is normalized to obtain an attention weight between two wind and light stations; The attention score represents the influence between different wind and light stations. In this embodiment, the normalization is performed by a Softmax function. Taking three wind and light stations A, B and C as an example, there are AB attention score and AC attention score for the A station. After the normalization, there are AB attention weight and AC attention weight, and the sum of the two is 1.

[0034] The attention weights are combined into the dynamic correlation matrix. It can be understood that the dynamic correlation matrix is a set of attention weights between all wind and light stations. It can simultaneously predict the data of different wind and light stations.

[0035] S350: converting the input data set into a stage feature set through the dynamic correlation matrix, the stage feature set including stage sub-features under a plurality of time sequences; Specifically, for each wind and light station, on the basis of the original data, the original data of other wind and light stations at the corresponding time node is weighted by the attention weight, thereby forming the stage sub-feature. It can be understood that the number of stage sub-features corresponds to the number of output values to be used, and it is a high-dimensional feature formed by comprehensively considering the influence of output values to be used between different stations and the influence of meteorological sub-data.

[0036] S360: setting a split window in the time attention module through the time sequence length parameter, inputting the stage feature set into the time attention module to select a to-be-updated sub-feature from a plurality of stage sub-features, and selecting a plurality of reference sub-features from a plurality of stage sub-features based on the split window, generating a final sub-feature corresponding to the to-be-updated sub-feature through the to-be-updated sub-feature and the plurality of reference sub-features; The time attention module considers the time correlation of stage sub-features under different time frames, so that the final sub-feature obtained comprehensively considers the spatial relationship and historical change rule between different wind and light stations, further improving the accuracy of subsequent prediction.

[0037] Assuming that there are 5 stage sub-features of the stage, which are feature 1, feature 2, feature 3, feature 4 and feature 5, and the prediction needs to be based on feature 5, in the case that the time series length parameter is 3 and the length of the split window is 3, the sub-feature to be updated is feature 5, and the reference sub-features are feature 3 and feature 4. After obtaining the reference sub-features and the sub-feature to be updated, the three are weighted by the time attention weight matrix to obtain the final sub-feature. Understandably, when processing the reference sub-features and the sub-feature to be updated, the time attention module will generate a hidden state vector at each time step, which condenses the historical information up to the current time. The correlation score between the current time hidden state vector and the historical time hidden state vector is calculated, and then the time attention weight matrix is formed through normalization processing. In this embodiment, the correlation score is obtained by a score function.

[0038] S370: Combine the final sub-features into a prediction output set, and construct a loss function by the prediction output set and the result data set to update the initial output prediction model to a final output prediction model. Understandably, because different wind and light stations need to be simultaneously predicted, the final sub-features are the predicted values of different stations at different time frames. By constructing a loss function with the recorded values of different stations at different time frames in the result data set, the initial output prediction model can be iteratively optimized by the loss function.

[0039] S40: Obtain a prediction error corresponding to the historical predicted output data, obtain real predicted output data by the final output prediction model, expand the real predicted output data into a real predicted output range by the prediction errors, and compare the real output data with the real predicted output range to determine whether there is a maintenance requirement. Because the output of wind and light stations has uncertainty such as intermittency, randomness and volatility, the uncertainty caused by the prediction error cannot be estimated by only the deterministic point prediction value. In order to balance the prediction accuracy and risk control, the uncertainty needs to be quantified to a certain extent. The historical predicted output data is the prediction output set, and the difference is only the result obtained by different stages of prediction models. By comparing the historical predicted output data with the result data set, the prediction errors are formed. The real predicted output data is the result value predicted by the final output prediction model according to the output data at the current time.

[0040] The step S40 comprises: S410: a plurality of the prediction errors are sorted, and a first position threshold and a second position threshold are set; In the embodiment, the first position threshold is 0.025, and the second position threshold is 0.975.

[0041] S420: the total number of a plurality of the prediction errors is obtained, a first error is selected from a plurality of the prediction errors through the total number and the first position threshold, and a second error is selected from a plurality of the prediction errors through the total number and the second position threshold; Understandably, if there are 2000 prediction errors, the first error is the 50th prediction error, and the second error is the 1950th prediction error.

[0042] S430: the real prediction output data is superimposed with the first error and the second error respectively to form a real prediction output range.

[0043] Understandably, if the real output data is within the real prediction output range, the corresponding wind and light field station does not need to be maintained, and if the real output data is outside the real prediction output range, the corresponding wind and light field station needs to be maintained.

[0044] By obtaining the original output data and the original weather data between different wind and light field stations, the time series length parameter and the spatial correlation matrix are constructed, which breaks through the limitation of traditional independent monitoring of a single station, quantifies the correlation between output fluctuations of different wind and light field stations and the periodicity of the output of the same wind and light field station, and improves the accuracy of output prediction; by generating the real prediction output range based on the prediction error, the output uncertainty of the wind and light field station is quantified, a reasonable fluctuation range of output prediction is set, and the false alarm rate of maintenance is reduced; by constructing the final output prediction model, based on the foregoing method, end-to-end automated remote intelligent operation and maintenance is realized, a means for real-time monitoring of wind and light field stations is provided, the consumption of manpower and material resources is reduced, time cost is saved, and the real-time and accuracy of maintenance are ensured.

[0045] Please refer to Figure 2 The second embodiment of the present application provides a remote intelligent operation and maintenance system based on photovoltaic and wind power data, which is applied to the remote intelligent operation and maintenance method based on photovoltaic and wind power data in the above embodiment, which has been described and will not be repeated. As used below, the terms "module", "unit", "subunit", etc. can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware or a combination of software and hardware is also possible and is conceived.

[0046] The system comprises: a processing module 10, configured to acquire a plurality of original data sets of a plurality of wind and light stations, the original data sets comprising original output data and original meteorological data, and to preprocess the original data sets to acquire a to-be-used data set comprising to-be-used output data and to-be-used meteorological data; The processing module 10 comprises: a first unit, configured to perform missing data filling and abnormality processing on the original output data and the original meteorological data to acquire to-be-used output data and stage meteorological data; a second unit, configured to perform normalization processing on the stage meteorological data to acquire to-be-used meteorological data, the to-be-used output data and the to-be-used meteorological data constituting the to-be-used data set; an extraction module 20, configured to acquire, by a plurality of the to-be-used output data, a time series length parameter corresponding to all wind and light stations and a spatial correlation matrix; The extraction module 20 comprises: a third unit, configured to acquire a station output mean value based on a plurality of the to-be-used output values, to set a plurality of delay orders, and to acquire, by the to-be-used output values, the station output mean value and the delay orders, a time series correlation coefficient corresponding to the delay orders; a fourth unit, configured to select, as a candidate correlation coefficient, a time series correlation coefficient greater than a first correlation threshold value, and to select, as a time series length parameter, a delay order corresponding to the smallest candidate correlation coefficient; a fifth unit, configured to acquire, by the to-be-used output values and the station output mean value, a spatial correlation coefficient between two wind and light stations, to construct a correlation relationship between two wind and light stations with a spatial correlation coefficient greater than a second correlation threshold value, and to form a spatial correlation matrix; a prediction module 30, configured to construct an initial output prediction model, to construct an input data set and a result data set, to update the initial output prediction model to a final output prediction model by the input data set, the result data set, the time series length parameter and the spatial correlation matrix, and to acquire historical prediction output data by the final output prediction model; The prediction module 30 comprises: a sixth unit, configured to select a reference time node, and to split the to-be-used output data into a first data set and a result data set by the reference time node; a seventh unit, configured to split, from the to-be-used meteorological data, a second data set corresponding to the first data set by the reference time node; an eighth unit, configured to splice the first data set and the second data set into an input data set; The ninth unit is configured to construct a dynamic correlation matrix in the spatial convolution module based on the input data set and the spatial correlation matrix; The ninth unit is configured to obtain an attention score between two wind and light field stations based on the input data set, normalize the attention score to obtain an attention weight between the two wind and light field stations, and combine a plurality of attention weights into the dynamic correlation matrix. The tenth unit is configured to convert the input data set into a stage feature set based on the dynamic correlation matrix, wherein the stage feature set comprises stage sub-features in a plurality of time sequences. The eleventh unit is configured to set a segmentation window in the time attention module based on the time sequence length parameter, input the stage feature set into the time attention module, select a to-be-updated sub-feature from the plurality of stage sub-features, and generate a final sub-feature corresponding to the to-be-updated sub-feature based on the to-be-updated sub-feature and a plurality of reference sub-features selected from the plurality of stage sub-features based on the segmentation window. The twelfth unit is configured to combine a plurality of final sub-features into a prediction output set, construct a loss function based on the prediction output set and the result data set, and update the initial output prediction model into a final output prediction model. The execution module 40 is configured to obtain prediction errors corresponding to the historical prediction output data, obtain real prediction output data based on the final output prediction model, expand the real prediction output data into a real prediction output range based on a plurality of prediction errors, compare real output data with the real prediction output range, and determine whether maintenance is required. The execution module 40 comprises: The thirteenth unit is configured to sort a plurality of prediction errors and set a first position threshold and a second position threshold. The fourteenth unit is configured to obtain a total number of a plurality of prediction errors, select a first error from a plurality of prediction errors based on the total number and the first position threshold, and select a second error from a plurality of prediction errors based on the total number and the second position threshold. The fifteenth unit is configured to superimpose the real prediction output data with the first error and the second error, respectively, to form a real prediction output range.

[0047] The application further provides a computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the remote intelligent operation and maintenance method based on photovoltaic and wind power data when executing the computer program.

[0048] The application further provides a storage medium, which stores a computer program, and the computer program is executed by a processor to realize the remote intelligent operation and maintenance method based on photovoltaic and wind power data.

[0049] In the description of the present specification, the description of the terms “one embodiment”, “some embodiments”, “an example”, “a specific example”, or “some examples” and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0050] The above-described embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A remote intelligent operation and maintenance method based on photovoltaic and wind power data, characterized in that, Includes the following steps: Several raw datasets from several wind and solar power stations are obtained. The raw datasets include raw power output data and raw meteorological data. The raw datasets are preprocessed to obtain a ready dataset including ready power output data and ready meteorological data. The time series length parameters and spatial correlation matrix corresponding to all wind and solar power stations are obtained by using several of the available power output data. An initial power output prediction model is constructed, along with an input dataset and a result dataset. The initial power output prediction model is then updated to a final power output prediction model using the input dataset, the result dataset, the time series length parameter, and the spatial correlation matrix. Historical predicted power output data is then obtained using the final power output prediction model. Obtain the prediction error corresponding to the historical predicted output data, obtain the actual predicted output data through the final output prediction model, expand the actual predicted output data into an actual predicted output range through several prediction errors, and compare the actual output data with the actual predicted output range to determine whether there is a maintenance requirement.

2. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 1, characterized in that, The step of preprocessing the original dataset to obtain a pending dataset including pending power output data and pending meteorological data includes: The raw power output data and raw meteorological data are filled with missing data and anomalies are processed to obtain the power output data to be used and the meteorological data of the stage. The meteorological data of the aforementioned stage are normalized to obtain standby meteorological data. The standby power output data and the standby meteorological data constitute the standby dataset.

3. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 1, characterized in that, The available power output data includes available power output values ​​in several consecutive time frames. The step of obtaining the time series length parameter and spatial correlation matrix corresponding to all wind and solar power stations through several of the available power output data includes: The average power output of the station is obtained based on several available power output values. Several delay orders are set, and the time-series correlation coefficient corresponding to the delay order is obtained through the available power output values, the average power output of the station, and the delay order. The time series correlation coefficients that are greater than the first correlation threshold are selected as candidate correlation coefficients, and the delay order corresponding to the smallest candidate correlation coefficient is selected as the time series length parameter. The spatial correlation coefficient between two wind and solar power stations is obtained by using the available power output value and the average power output of the stations. A correlation relationship is constructed between two wind and solar power stations whose spatial correlation coefficient is greater than the second correlation threshold to form a spatial correlation matrix.

4. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 3, characterized in that, The formula for obtaining the time-series correlation coefficient is: ， in, Represents the time-series correlation coefficient. This represents the available output value in the t-th time frame. This represents the average power output of the power station. Indicates the delay order. Indicates the total sampling time; The formula for obtaining the spatial correlation coefficient is: ， in, This represents the spatial correlation coefficient between the i-th and j-th wind and solar power stations. This represents the available power output value of the i-th wind farm station in the t-th time frame. This represents the standby power output value of the j-th wind farm station in the t-th time frame. Let represent the average power output of the i-th wind and solar power station. This represents the average power output of the j-th wind and solar power station.

5. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 1, characterized in that, The steps for constructing the input dataset and the result dataset include: A reference time node is selected, and the power output data to be used is divided into a first dataset and a result dataset based on the reference time node; A second dataset corresponding to the first dataset is extracted from the meteorological data to be used based on the reference time node; The first dataset and the second dataset are concatenated to form the input dataset.

6. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 1, characterized in that, The initial power output prediction model includes a spatial convolution module and a temporal attention module. The step of updating the initial power output prediction model to the final power output prediction model using the input dataset, the result dataset, the time series length parameter, and the spatial correlation matrix includes: Based on the spatial correlation matrix, a dynamic correlation matrix is ​​constructed in the spatial convolution module using the input dataset; The input dataset is converted into a stage feature set through the dynamic correlation matrix, and the stage feature set includes several stage sub-features under time series. The time series length parameter is used to set a segmentation window in the time attention module. The stage feature set is input into the time attention module to select the sub-feature to be updated from several stage sub-features. Based on the segmentation window, several reference sub-features are selected from several stage sub-features. The final sub-feature corresponding to the sub-feature to be updated is generated through the sub-feature to be updated and several reference sub-features. Several of the final sub-features are combined into a prediction output set, and a loss function is constructed using the prediction output set and the result dataset to update the initial output prediction model into a final output prediction model.

7. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 6, characterized in that, The step of constructing a dynamic correlation matrix in the spatial convolution module using the input dataset includes: The attention score between the two wind and solar power stations is obtained using the input dataset. The attention scores are normalized to obtain the attention weights between the two wind and solar power stations. The attention weights are combined into the dynamic correlation matrix.

8. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 7, characterized in that, The formula for obtaining the attention score is: ， in, This represents the attention score between the i-th scenic spot and the j-th scenic spot at the t-th time frame. , , , , All of these represent learnable parameters. Indicates the transpose operator. This represents the time frame sub-data of the i-th and j-th wind and solar power stations in the input dataset at time frame t. This represents the activation function.

9. The remote intelligent operation and maintenance method based on photovoltaic and wind power data according to claim 1, characterized in that, The step of expanding the actual predicted output data to the actual predicted output range using a plurality of prediction errors includes: The prediction errors are sorted and a first position threshold and a second position threshold are set. Obtain the total number of several prediction errors, select a first error from several prediction errors using the total number and the first position threshold, and select a second error from several prediction errors using the total number and the second position threshold; The actual predicted output data is superimposed with the first error and the second error respectively to form the actual predicted output range.

10. A remote intelligent operation and maintenance system based on photovoltaic and wind power data, applied to the remote intelligent operation and maintenance method based on photovoltaic and wind power data as described in any one of claims 1 to 9, characterized in that, The system includes: The processing module is used to acquire several raw datasets from several wind and solar power stations. The raw datasets include raw power output data and raw meteorological data. The raw datasets are preprocessed to obtain a ready dataset including ready power output data and ready meteorological data. The extraction module is used to obtain the time series length parameter and spatial correlation matrix corresponding to all wind and solar power stations from several of the power output data to be used; The prediction module is used to construct an initial power output prediction model, and to construct an input dataset and a result dataset. The initial power output prediction model is updated to a final power output prediction model using the input dataset, the result dataset, the time series length parameter, and the spatial correlation matrix. Historical predicted power output data is obtained through the final power output prediction model. The execution module is used to obtain the prediction error corresponding to the historical predicted output data, obtain the actual predicted output data through the final output prediction model, expand the actual predicted output data into an actual predicted output range through several prediction errors, and compare the actual output data with the actual predicted output range to determine whether there is a maintenance requirement.

Images