CN114065667A - SF6 equipment gas pressure prediction method based on Prophet-LSTM model - Google Patents

SF6 equipment gas pressure prediction method based on Prophet-LSTM model Download PDF

Info

Publication number
CN114065667A
CN114065667A CN202111391722.0A CN202111391722A CN114065667A CN 114065667 A CN114065667 A CN 114065667A CN 202111391722 A CN202111391722 A CN 202111391722A CN 114065667 A CN114065667 A CN 114065667A
Authority
CN
China
Prior art keywords
model
data
prophet
prediction
lstm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111391722.0A
Other languages
Chinese (zh)
Other versions
CN114065667B (en
Inventor
陆敏安
郑真
陈敬德
顾华
樊汝森
黄强
陈亚杰
徐友刚
张红燕
李建宁
黄一楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Shanghai Electric Power Co Ltd
Original Assignee
State Grid Shanghai Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by State Grid Shanghai Electric Power Co Ltd filed Critical State Grid Shanghai Electric Power Co Ltd
Priority to CN202111391722.0A priority Critical patent/CN114065667B/en
Publication of CN114065667A publication Critical patent/CN114065667A/en
Application granted granted Critical
Publication of CN114065667B publication Critical patent/CN114065667B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/28Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/044Recurrent networks, e.g. Hopfield networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/047Probabilistic or stochastic networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00Details relating to the application field
    • G06F2113/08Fluids
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14Force analysis or force optimisation, e.g. static or dynamic forces

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Computer Hardware Design (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Geometry (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Algebra (AREA)
  • Fluid Mechanics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a gas pressure prediction method of SF6 equipment based on a Prophet-LSTM model, which comprises the following steps: collecting data of SF6 electrical equipment during operation, preprocessing the data, and dividing the data into a training set and a test set after normalization; training the data of the training set respectively through a Prophet model and an LSTM (long short term memory network) model optimized by Bayes; inputting the prediction results of the training set of the Prophet model and the LSTM model and the real data of the training set into an optimal weight coefficient acquisition module to obtain the combined weight of the two models; inputting test set data to the combined model, and checking the prediction performance of the combined model according to the test set prediction result and the test set real data; finally, the perfect combined prediction model is used for SF6 pressure prediction. The Prophet-LSTM model is applied to prediction of the pressure value in SF6 electrical equipment, so that the method is beneficial to judging possible overheating faults and leakage faults in the equipment through the gas pressure value, and plays a role in early warning.

Description

SF6 equipment gas pressure prediction method based on Prophet-LSTM model
Technical Field
The invention relates to a gas pressure prediction method of SF6 equipment based on a Prophet-LSTM model, which is used in the field of power equipment detection.
Background
SF6 (sulfur hexafluoride) gas has good insulation and arc extinguishing characteristics, and is widely applied to power transformation equipment at present, however, SF6 has some hidden troubles as an insulation medium of the power transformation equipment. For example, in the gas leakage problem generally existing in the closed GIS (gas insulated substation) equipment, once leakage occurs, due to the effect of a high-voltage arc, SF6 is decomposed to generate some toxic substances, so that the air in a GIS room is anoxic and toxic, and the safety of personnel is threatened, and on the other hand, SF6 gas leakage also reduces the insulation and arc extinguishing performance of the equipment, so that measures are necessary to be taken to realize the rapid and effective monitoring of SF6 gas pressure, and thus the leakage fault of the equipment is judged. Besides, most faults of the primary equipment of the power transformation are expressed in the form of heat, the heat is released through SF6 gas, the SF6 gas pressure is inevitably changed in the heat exchange process, and therefore real-time monitoring of the SF6 gas pressure is also of great significance for analyzing overheating faults of the equipment.
At present, for the prediction of time series data such as SF6 gas pressure, a single time series model or a prediction method based on a neural network is mostly adopted, however, time series generally comprises linear components and nonlinear components, the single prediction method can only guarantee the prediction accuracy of one of the components, and in addition, the single prediction model has a defect in effectively capturing the composite characteristics of the time series.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a gas pressure prediction method of SF6 equipment based on a Prophet-LSTM model, and by applying a combined prediction model, the advantages of a time series model and a prediction method based on a neural network are fully exerted, and compared with a single model, the prediction method has a better prediction effect. The time sequence models of Prophet and LSTM optimized by Bayes are used for predicting SF6 air pressure time sequence data respectively, and the prediction results of the two models are subjected to linear weighting to serve as the final SF6 air pressure prediction result, so that SF6 air pressure change can be predicted more accurately and effectively.
One technical scheme for achieving the above purpose is as follows: a gas pressure prediction method of SF6 equipment based on a Prophet-LSTM model comprises the following steps:
s1, acquiring SF6 air pressure time sequence data in the past year through an air pressure sensor of SF6 equipment;
s2, preprocessing the acquired SF6 historical air pressure data: the method comprises the steps of processing missing data, repeated and redundant data, abnormal or wrong data, further normalizing the data in consideration of the problem that the value range difference of the data in a training model is overlarge and possibly causes the wrong weight distribution of the model, and then dividing the data into a training set and a test set;
s3, training: respectively inputting the training set data into a Prophet model and an LSTM model optimized by Bayes for training, and simultaneously obtaining the optimal weight coefficient of the combined model;
s4, testing: after the training of the Prophet-LSTM combined model is finished, inputting the test set data into the Prophet-LSTM combined model for prediction, and taking the linear weighted value of the output results of the Prophet-LSTM combined model and the Prophet-LSTM combined model as the final test set prediction result;
and S5, testing the prediction effect of the combined model, if the error between the prediction result of the test set and the actual test set meets the requirement, keeping the structure and parameters of the current combined prediction model, and if not, returning to the training stage.
Further, the step S2 specifically includes:
s21, processing missing data by adopting a proximity value filling method, and replacing the missing value by the average value of the front and back adjacent position data;
s22, deleting repeated or redundant data;
s23, recognizing abnormal and error data by using a manually set range, and processing the abnormal and error data as a missing value;
s24, in order to accelerate the convergence rate and the prediction precision of the prediction model, the data is further processed by adopting a minimum-maximum (min-max) normalization method, the data is compressed to an interval [0,1], and meanwhile, the influence of different dimensions is eliminated, and the conversion formula is as follows:
Figure BDA0003364548840000031
wherein x is the actual value of the historical pressure data of SF6, max and min are the maximum value and the minimum value respectively, and x*The normalized data is represented.
Further, the Prophet model in step S3 can effectively analyze the time series characteristics of the data, such as periodicity, holiday effect, and future trend, and the steps of training with the Prophet model are as follows:
the Prophet model decomposes the equipment data into three parts, g (t) represents an aperiodic term in the data, h (t) represents a holiday term in the data, and s (t) represents a periodic term in the data, and the formula is as follows:
P=g(t)+s(t)+h(t)+εt
for data of non-periodic terms, modeling is carried out by using a logic growth model, and the formula is as follows:
Figure BDA0003364548840000032
wherein C represents the capacity of the prediction model, k is the trend growth rate, and n represents the offset parameter;
the periodic effect is simulated using a fourier series for the data of the holiday and periodic terms, the formula being:
Figure BDA0003364548840000033
and determining parameters in the final model through testing to obtain a Prophet model predicted value P.
Further, the LSTM model in step S3 alleviates the problems of gradient extinction and gradient explosion when the recurrent neural network processes long-sequence data by adding forgetting gates to the neuron part, and is suitable for training long-distance time-series data, and the LSTM model is implemented as follows:
s41, initializing the weight and the offset parameter of the neural network, and then searching the optimal hyper-parameter combination of the LSTM model through a Bayesian optimization method, wherein the Bayesian optimization does not need derivation, thereby being suitable for adjusting the hyper-parameter of the neural network and finding a better hyper-parameter combination with fewer steps;
and S42, inputting training set data to train the LSTM neural network.
Further, in step S3, the training set data are respectively input to the Prophet model and the LSTM model optimized by bayes, the obtained training set prediction results are respectively represented as p (t) and l (t), the two are linearly weighted and then input to the optimal weight coefficient obtaining module together with the training set real data, and the training set prediction results obtained by combining the models are as follows:
R(t)=ω1 *P(t)+ω2 *L(t)
in the formula, ω1 *2 *When the training set prediction result r (t) is closest to the actual training set, the weight coefficient ω is 11 *And ω2 *The value of (a) is determined by the least squares method.
Further, in step S4, the test set data is input into the combined model, and a final test set prediction result is obtained according to the prediction results of the two models and the optimal combined weight.
Further, the step S5 is implemented as follows:
the two indexes of average absolute percentage and root mean square error are adopted to test the prediction effect of the combined model, and the formulas are respectively as follows:
Figure BDA0003364548840000041
Figure BDA0003364548840000042
wherein p isiAnd aiAnd respectively representing the prediction result of the combined model at the moment i on the test data and the true value of the test data, if the MAPE and RMSE indexes meet the requirements, storing the structure and parameters of the current combined model, and otherwise, continuing to train the model.
Compared with the prior art, the invention has the following advantages:
1. according to the invention, the Prophet-LSTM combined model is adopted to predict SF6 air pressure time sequence data, compared with a single prediction model, the characteristics of time sequence data can be better analyzed, and the accuracy is higher when complex nonlinear time sequences are predicted;
2. the LSTM model adopted by the invention can effectively improve the problems of gradient explosion and disappearance of the simple circulating neural network, and meanwhile, the LSTM model is optimized by using a Bayesian method, so that the optimal hyper-parameter combination of the neural network is searched in a faster and simpler way, and the training speed and the prediction accuracy of the neural network are improved.
Drawings
Fig. 1 is a flow chart of the SF6 plant gas pressure prediction method of the present invention.
Detailed Description
In order to better understand the technical solution of the present invention, the following detailed description is made by specific examples:
referring to fig. 1, the SF6 equipment gas pressure prediction method based on the Prophet-LSTM model of the present invention includes the following steps:
s1, acquiring SF6 air pressure time sequence data in the past year through an air pressure sensor of SF6 equipment;
s2, preprocessing the acquired SF6 historical air pressure data: the method comprises the steps of processing missing data, repeated and redundant data, abnormal or wrong data, further normalizing the data in consideration of the problem that the value range difference of the data in a training model is overlarge and possibly causes the wrong weight distribution of the model, and then dividing the data into a training set and a test set;
s3, training: respectively inputting the training set data into a Prophet model and an LSTM model optimized by Bayes for training, and simultaneously obtaining the optimal weight coefficient of the combined model;
s4, testing: after the training of the Prophet-LSTM combined model is finished, inputting the test set data into the Prophet-LSTM combined model for prediction, and taking the linear weighted value of the output results of the Prophet-LSTM combined model and the Prophet-LSTM combined model as the final test set prediction result;
and S5, testing the prediction effect of the combined model, if the error between the prediction result of the test set and the actual test set meets the requirement, keeping the structure and parameters of the current combined prediction model, and if not, returning to the training stage.
The step S2 is specifically implemented as follows:
s21, processing missing data by adopting a proximity value filling method, and replacing the missing value by the average value of the front and back adjacent position data;
s22, deleting repeated or redundant data;
s23, recognizing abnormal and error data by using a manually set range, and processing the abnormal and error data as a missing value;
s24, in order to accelerate the convergence rate and the prediction precision of the prediction model, the data is further processed by adopting a minimum-maximum (min-max) normalization method, the data is compressed to an interval [0,1], and meanwhile, the influence of different dimensions is eliminated, and the conversion formula is as follows:
Figure BDA0003364548840000061
wherein x is the actual value of the historical pressure data of SF6, max and min are the maximum value and the minimum value respectively, and x*The normalized data is represented.
The Prophet model in the step S3 can effectively analyze the time series characteristics of the data, such as periodicity, holiday effect, future trend, and the like, and the steps of training by using the Prophet model are as follows:
the Prophet model decomposes the equipment data into three parts, g (t) represents an aperiodic term in the data, h (t) represents a holiday term in the data, and s (t) represents a periodic term in the data, and the formula is as follows:
P=g(t)+s(t)+(t)+εt
for data of non-periodic terms, modeling is carried out by using a logic growth model, and the formula is as follows:
Figure BDA0003364548840000062
where C represents the capacity of the prediction model, k is the trend growth rate, and n represents the offset parameter.
The periodic effect is simulated using a fourier series for the data of the holiday and periodic terms, the formula being:
Figure BDA0003364548840000063
and determining parameters in the final model through testing to obtain a Prophet model predicted value P.
The LSTM model in step S3 effectively alleviates the problems of gradient extinction and gradient explosion when the Recurrent Neural Network (RNN) processes long-sequence data by adding forgetting gates to the neuron portion, and is suitable for training long-distance time-series data, and the LSTM model is implemented as follows:
s41, initializing the weight and the offset parameter of the neural network, and then searching the optimal hyper-parameter combination of the LSTM model through a Bayesian optimization method, wherein the Bayesian optimization does not need derivation, thereby being suitable for adjusting the hyper-parameter of the neural network and finding a better hyper-parameter combination with fewer steps;
and S42, inputting training set data to train the LSTM neural network.
In step S3, the training set data are respectively input to the Prophet model and the LSTM model optimized by bayes, the obtained training set prediction results are respectively expressed as p (t) and l (t), the two are linearly weighted and then input to the optimal weight coefficient obtaining module together with the training set real data, and the training set prediction results obtained by combining the models are as follows:
R(t)=ω1 *P(t)+ω2 *L(t)
in the formula, ω1 *2 *When the training set prediction result r (t) is closest to the actual training set, the weight coefficient ω is 11 *And ω2 *The value of (a) is determined by the least squares method.
And step S4, inputting the test set data into the combined model, and obtaining a final test set prediction result according to the prediction results of the two models and the optimal combined weight.
The step S5 is specifically implemented as follows:
the two indexes of the Mean Absolute Percentage (MAPE) and the Root Mean Square Error (RMSE) are adopted to test the prediction effect of the combined model, and the formulas are respectively as follows:
Figure BDA0003364548840000071
Figure BDA0003364548840000072
wherein p isiAnd aiAnd respectively representing the prediction result of the combined model at the moment i on the test data and the true value of the test data, if the MAPE and RMSE indexes meet the requirements, storing the structure and parameters of the current combined model, and otherwise, continuing to train the model.
It should be understood by those skilled in the art that the above embodiments are only for illustrating the present invention and are not to be used as a limitation of the present invention, and that changes and modifications to the above described embodiments are within the scope of the claims of the present invention as long as they are within the spirit and scope of the present invention.

Claims (7)

1.一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,包括如下步骤:1. a SF6 equipment gas pressure prediction method based on Prophet-LSTM model, is characterized in that, comprises the steps: S1、通过SF6设备的气压传感器获取过去一年中SF6气压时序数据;S1. Obtain the SF6 air pressure time series data in the past year through the air pressure sensor of the SF6 equipment; S2、预处理获取的SF6历史气压数据:包括对缺失数据、重复和冗余数据、异常或错误数据的处理,考虑到训练模型中数据取值范围相差过大可能引起模型权重分配错误的问题,进一步对数据进行归一化处理,然后将其分为训练集和测试集;S2. The SF6 historical air pressure data obtained by preprocessing: including the processing of missing data, repeated and redundant data, abnormal or wrong data, considering that the large difference in the data value range in the training model may cause the problem of wrong model weight allocation, Further normalize the data, and then divide it into training set and test set; S3、训练阶段:将训练集数据分别输入到Prophet模型和经贝叶斯优化的LSTM模型进行训练,同时获取组合模型的最优权重系数;S3. Training stage: input the training set data into the Prophet model and the Bayesian optimized LSTM model respectively for training, and obtain the optimal weight coefficient of the combined model at the same time; S4、测试阶段:待Prophet-LSTM组合模型训练完成后,将测试集数据输入到Prophet-LSTM组合模型进行预测,以二者输出结果的线性加权值为最终的测试集预测结果;S4. Test stage: After the training of the Prophet-LSTM combined model is completed, the test set data is input into the Prophet-LSTM combined model for prediction, and the linear weighted value of the output results of the two is used as the final test set prediction result; S5、对组合模型的预测效果进行检验,若测试集预测结果与实际的测试集误差满足要求,则保留当前的组合预测模型的结构与参数,否则返回训练阶段。S5. Test the prediction effect of the combined model. If the error between the prediction result of the test set and the actual test set meets the requirements, keep the structure and parameters of the current combined prediction model, otherwise return to the training stage. 2.根据权利要求1所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S2的具体包括:2. a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, is characterized in that, described step S2 specifically comprises: S21、采用邻近值填补方法处理缺失数据,以前后相邻位置数据的平均值代替缺失值;S21. Use the adjacent value filling method to process the missing data, and replace the missing value with the average value of the data of the adjacent position before and after; S22、删除重复或冗余的数据;S22. Delete duplicate or redundant data; S23、利用人为设定的范围来识别异常和错误数据,并将其视为缺失值进行处理;S23. Use artificially set ranges to identify abnormal and erroneous data, and treat them as missing values; S24、为了加快预测模型的收敛速度以及预测精度,采用最小最大(min-max)归一化方法对数据做进一步处理,将数据压缩至区间[0,1],同时消除量纲不同的影响,转换公式如下:S24. In order to speed up the convergence speed and prediction accuracy of the prediction model, the min-max normalization method is used to further process the data, compress the data to the interval [0,1], and eliminate the influence of different dimensions. The conversion formula is as follows:
Figure FDA0003364548830000011
Figure FDA0003364548830000011
其中,x为SF6历史气压数据的实际值,max和min分别为其最大值和最小值,x*表示归一化后的数据。Among them, x is the actual value of the SF6 historical air pressure data, max and min are the maximum and minimum values, respectively, and x * represents the normalized data.
3.根据权利要求1所述的所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S3中Prophet模型能够有效分析数据的周期性、节假日效应以及未来趋势等时间序列特征,使用Prophet模型进行训练的步骤如下:3. the described a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, is characterized in that, in described step S3, Prophet model can effectively analyze the periodicity of data, holiday effect and future For time series features such as trends, the steps to use the Prophet model for training are as follows: Prophet模型将设备数据分解为三个部分,g(t)表示数据中的非周期项、h(t)表示数据中的节假日项,s(t)表示数据中的周期项,公式为:The Prophet model decomposes the equipment data into three parts, g(t) represents the aperiodic item in the data, h(t) represents the holiday item in the data, and s(t) represents the periodic item in the data. The formula is: P=g(t)+s(t)+h(t)+εt P=g(t)+s(t)+h(t)+ε t 对于非周期项的数据,使用逻辑增长模型进行建模,公式为:For the data of aperiodic items, use the logistic growth model to model, the formula is:
Figure FDA0003364548830000021
Figure FDA0003364548830000021
其中,C表示预测模型的容量,k是趋势增长率,n表示偏移参数;Among them, C represents the capacity of the prediction model, k is the trend growth rate, and n represents the offset parameter; 对于节假日和周期项的数据使用傅里叶级数来模拟周期效应,公式为:Use Fourier series to simulate the periodic effect for the holiday and periodic term data, the formula is:
Figure FDA0003364548830000022
Figure FDA0003364548830000022
其中T为周期性时间序列的周期,通过测试确定最终模型中的参数,得到Prophet模型预测值P。Where T is the period of the periodic time series, and the parameters in the final model are determined by testing, and the predicted value P of the Prophet model is obtained.
4.根据权利要求1所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S3中LSTM模型通过在神经元部分增加忘记门缓解循环神经网络处理长序列数据时存在的梯度消失和梯度爆炸问题,适合训练长距离的时序数据,其具体实现如下:4. a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, is characterized in that, in described step S3, LSTM model alleviates cyclic neural network processing long sequence by adding forget gate in neuron part The gradient disappearance and gradient explosion problems existing in the data are suitable for training long-distance time series data. The specific implementation is as follows: S41、初始化神经网络的权重和偏置参数,然后通过贝叶斯优化方法寻找LSTM模型的最优超参数组合,贝叶斯优化不需要求导,因而适用于调节神经网络的超参数,并且能以较少的步数找到比较好的超参数组合;S41. Initialize the weight and bias parameters of the neural network, and then use the Bayesian optimization method to find the optimal hyperparameter combination of the LSTM model. Bayesian optimization does not require derivation, so it is suitable for adjusting the hyperparameters of the neural network, and can Find a better combination of hyperparameters with fewer steps; S42、输入训练集数据对LSTM神经网络进行训练。S42, input the training set data to train the LSTM neural network. 5.根据权利要求1所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S3中将训练集数据分别输入到Prophet模型和经贝叶斯优化的LSTM模型,得到的训练集预测结果分别表示为P(t)和L(t),将二者进行线性加权后与训练集真实数据一起输入到最优权重系数获取模块,组合模型最终得到的训练集预测结果如下:5. a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, is characterized in that, in described step S3, input the training set data to Prophet model and the LSTM optimized by Bayesian respectively model, the obtained training set prediction results are expressed as P(t) and L(t), respectively, after linearly weighting the two, together with the real data of the training set, they are input to the optimal weight coefficient acquisition module, and the final training set obtained by combining the model The predicted results are as follows: R(t)=ω1 *P(t)+ω2 *L(t)R(t)=ω 1 * P(t)+ω 2 * L(t) 式中,ω1 *2 *=1,此时训练集预测结果R(t)与实际的训练集最接近,权重系数ω1 *和ω2 *的取值通过最小二乘法来确定。In the formula, ω 1 *2 * =1, at this time, the training set prediction result R(t) is the closest to the actual training set, and the values of the weight coefficients ω 1 * and ω 2 * are determined by the least square method. 6.根据权利要求1所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S4将测试集数据输入至组合模型中,根据两个模型的预测结果及最优的组合权重得到最终的测试集预测结果。6. a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, it is characterized in that, described step S4 input test set data in combined model, according to the prediction result of two models and The optimal combined weights get the final test set prediction results. 7.根据权利要求1所述的一种基于Prophet-LSTM模型的SF6设备气体压力预测方法,其特征在于,所述步骤S5的具体实现如下:7. a kind of SF6 equipment gas pressure prediction method based on Prophet-LSTM model according to claim 1, is characterized in that, the concrete realization of described step S5 is as follows: 采用平均绝对百分比以及均方根误差这两个指标对组合模型的预测效果进行检验,公式分别如下:The prediction effect of the combined model is tested by using the average absolute percentage and the root mean square error. The formulas are as follows:
Figure FDA0003364548830000031
Figure FDA0003364548830000031
Figure FDA0003364548830000032
Figure FDA0003364548830000032
其中pi和ai分别代表时刻i组合模型对测试数据的预测结果与测试数据的真实值,若MAPE和RMSE指标满足要求,则保存当前组合模型的结构与参数,否则继续对模型进行训练。Among them, p i and a i respectively represent the prediction result of the combined model at time i and the real value of the test data. If the MAPE and RMSE indicators meet the requirements, the structure and parameters of the current combined model will be saved, otherwise, the model will continue to be trained.
CN202111391722.0A 2021-11-19 2021-11-19 SF6 equipment gas pressure prediction method based on Prophet-LSTM model Active CN114065667B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111391722.0A CN114065667B (en) 2021-11-19 2021-11-19 SF6 equipment gas pressure prediction method based on Prophet-LSTM model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111391722.0A CN114065667B (en) 2021-11-19 2021-11-19 SF6 equipment gas pressure prediction method based on Prophet-LSTM model

Publications (2)

Publication Number Publication Date
CN114065667A true CN114065667A (en) 2022-02-18
CN114065667B CN114065667B (en) 2022-09-13

Family

ID=80279648

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111391722.0A Active CN114065667B (en) 2021-11-19 2021-11-19 SF6 equipment gas pressure prediction method based on Prophet-LSTM model

Country Status (1)

Country Link
CN (1) CN114065667B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114756604A (en) * 2022-06-13 2022-07-15 西南交通大学 A prediction method of monitoring time series data based on Prophet combination model
CN114925931A (en) * 2022-06-10 2022-08-19 北京中恒博瑞数字电力科技有限公司 Platform area load prediction method and system
CN115563475A (en) * 2022-10-25 2023-01-03 南京工业大学 Pressure soft sensor of excavator hydraulic system
CN115829101A (en) * 2022-11-21 2023-03-21 国网甘肃省电力公司酒泉供电公司 Electricity consumption cost prediction and optimization method based on multi-scale time convolution network
CN117370330A (en) * 2023-12-07 2024-01-09 浙江有数数智科技有限公司 Prophet-SVM model-based data monitoring and early warning method of big data acquisition system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990556A (en) * 2021-02-24 2021-06-18 江苏大学 User power consumption prediction method based on Prophet-LSTM model
CN113065703A (en) * 2021-03-31 2021-07-02 天津大学 Time series prediction method combining multiple models
CN113554148A (en) * 2021-06-07 2021-10-26 南京理工大学 A BiLSTM Voltage Bias Prediction Method Based on Bayesian Optimization

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990556A (en) * 2021-02-24 2021-06-18 江苏大学 User power consumption prediction method based on Prophet-LSTM model
CN113065703A (en) * 2021-03-31 2021-07-02 天津大学 Time series prediction method combining multiple models
CN113554148A (en) * 2021-06-07 2021-10-26 南京理工大学 A BiLSTM Voltage Bias Prediction Method Based on Bayesian Optimization

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
沈时宇等: "Prophet时序模型在短期水质溶氧预测中的应用", 《渔业现代化》 *
赵英等: "基于LSTM-Prophet非线性组合的时间序列预测模型", 《计算机与现代化》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114925931A (en) * 2022-06-10 2022-08-19 北京中恒博瑞数字电力科技有限公司 Platform area load prediction method and system
CN114756604A (en) * 2022-06-13 2022-07-15 西南交通大学 A prediction method of monitoring time series data based on Prophet combination model
CN114756604B (en) * 2022-06-13 2022-09-09 西南交通大学 A prediction method of monitoring time series data based on Prophet combination model
CN115563475A (en) * 2022-10-25 2023-01-03 南京工业大学 Pressure soft sensor of excavator hydraulic system
CN115829101A (en) * 2022-11-21 2023-03-21 国网甘肃省电力公司酒泉供电公司 Electricity consumption cost prediction and optimization method based on multi-scale time convolution network
CN115829101B (en) * 2022-11-21 2023-06-06 国网甘肃省电力公司酒泉供电公司 Power consumption cost prediction and optimization method based on multi-scale time convolution network
CN117370330A (en) * 2023-12-07 2024-01-09 浙江有数数智科技有限公司 Prophet-SVM model-based data monitoring and early warning method of big data acquisition system

Also Published As

Publication number Publication date
CN114065667B (en) 2022-09-13

Similar Documents

Publication Publication Date Title
CN114065667A (en) SF6 equipment gas pressure prediction method based on Prophet-LSTM model
CN111830408B (en) Motor fault diagnosis system and method based on edge calculation and deep learning
CN109726524B (en) CNN and LSTM-based rolling bearing residual service life prediction method
CN116610998A (en) Switch cabinet fault diagnosis method and system based on multi-mode data fusion
CN113988210B (en) Method, device and storage medium for repairing distorted data of structural monitoring sensor network
CN113516310A (en) Transformer fault early warning method and system
CN111122811A (en) A fault monitoring method for sewage treatment process based on OICA and RNN fusion model
CN112580198A (en) Improved optimization classification method for transformer state evaluation
CN111222992A (en) Stock price prediction method of long-short term memory neural network based on attention mechanism
Wang et al. A remaining useful life prediction model based on hybrid long-short sequences for engines
Fu et al. Research on gas outburst prediction model based on multiple strategy fusion improved snake optimization algorithm with temporal convolutional network
Zhang et al. Remaining useful life prediction for rolling bearings with a novel entropy-based health indicator and improved particle filter algorithm
Hu et al. A hybrid model for predicting the degradation trend of hydropower units based on deep learning
Tang et al. Prediction of bearing performance degradation with bottleneck feature based on LSTM network
CN117633690A (en) Rotary machine health state monitoring method and equipment based on data driving
CN114486262B (en) CNN-AT-LSTM-based bearing residual service life prediction method
CN114266201A (en) Self-attention elevator trapping prediction method based on deep learning
CN114487643A (en) On-spot handing-over of extra-high voltage GIL equipment is accepted and is synthesized test platform
Su et al. Prediction method for transformer state based on GRU network
Luo et al. A novel method for remaining useful life prediction of roller bearings involving the discrepancy and similarity of degradation trajectories
Wang et al. Failure rate prediction based on AR model and residual correction
CN113849479B (en) Oil tank leakage detection method for integrated energy supply station based on real-time learning and adaptive threshold
Huang et al. A prediction method for aero-engine health management based on nonlinear time series analysis
Zhang et al. Spatio-temporal fusion model of natural gas pipeline condition monitoring based on convolutional neural network and long short-term memory neural network
CN114861969A (en) A LSTM-based Power System Oscillation Risk Prediction Method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant