CN109816196A - Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model - Google Patents
Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model Download PDFInfo
- Publication number
- CN109816196A CN109816196A CN201811485623.7A CN201811485623A CN109816196A CN 109816196 A CN109816196 A CN 109816196A CN 201811485623 A CN201811485623 A CN 201811485623A CN 109816196 A CN109816196 A CN 109816196A
- Authority
- CN
- China
- Prior art keywords
- prediction model
- evaluation
- value
- true value
- prediction
- 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.)
- Pending
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of evaluation value calculation method of prediction model, device, equipment and readable storage medium storing program for executing, the method comprising the steps of: after getting training dataset, determine the true value that the training data is concentrated, predicted value corresponding with the true value is obtained by least one prediction model prestored, and calculates the error amount between the true value and the predicted value；The corresponding truthful data figure of the true value is drawn, and calculates the predictablity rate of focus in the truthful data figure；According to the evaluation of estimate of prediction model prediction result described in the error amount and the predictablity rate Calculation Estimation.The present invention is analyzed by data, using multiple evaluation index comprehensive automation valuation prediction models, improves the evaluation efficiency and evaluation accuracy of prediction model.
Description
Technical field
The present invention relates to data analysis technique field more particularly to a kind of evaluation value calculation method of prediction model, device,
Equipment and readable storage medium storing program for executing.
Background technique
Time series forecasting based on big data is the hot issue studied at present, while all kinds of prediction models continue to bring out,
To the evaluation of estimate counting system of prediction model, more stringent requirements are proposed.At present to the evaluation index list of prediction model prediction result
One, as determined predicted value quantity identical with true value in prediction result, by identical quantity divided by the total quantity institute of predicted value
The value obtained carrys out the prediction result of valuation prediction models, or using the prediction result of artificial valuation prediction models, so as to cause pre-
Survey model evaluation accuracy or evaluation inefficiency.
Summary of the invention
The main purpose of the present invention is to provide a kind of evaluation value calculation method of prediction model, device, equipment and readable
Storage medium, it is intended to solve existing prediction model evaluation efficiency and evaluate the low technical problem of accuracy.
To achieve the above object, the present invention provides a kind of evaluation value calculation method of prediction model, the prediction model
Evaluation value calculation method comprising steps of
After getting training dataset, determines the true value that the training data is concentrated, pass through at least one prestored
Prediction model obtains predicted value corresponding with the true value, and calculates the error amount between the true value and the predicted value；
The corresponding truthful data figure of the true value is drawn, and the prediction for calculating focus in the truthful data figure is accurate
Rate；
According to the evaluation of estimate of prediction model prediction result described in the error amount and the predictablity rate Calculation Estimation.
Preferably, described to calculate the truthful data when the focus is the wave crest point in the truthful data figure
The step of predictablity rate of focus, includes: in figure
Obtain the corresponding wave crest predicted value of the truthful data figure medium wave peak dot, the wave crest point corresponds to previous time period
The first interior true value and the corresponding wave crest true value of the wave crest point；
Calculate the first difference between the wave crest predicted value and first true value, and to calculate the wave crest true
The second difference between value and first true value；
The product for calculating first difference and second difference is greater than zero corresponding target wave crest point quantity, will be described
Target wave crest point quantity obtains the truthful data figure medium wave peak dot divided by all wave crest point quantity in the truthful data figure
Predictablity rate.
Preferably, described to calculate the truthful data when the focus is the turning point in the truthful data figure
The step of predictablity rate of focus, includes: in figure
It obtains the corresponding turnover predicted value of turning point, the turning point in the truthful data figure and corresponds to previous time period
The second interior true value and the corresponding turnover true value of the turning point；
The third difference between the turnover predicted value and second true value is calculated, and calculates the turnover really
The 4th difference between value and second true value；
The product for calculating the third difference and the 4th difference is greater than zero corresponding target inflection point quantity, will be described
Target inflection point quantity obtains turning point in the truthful data figure divided by the total quantity of turning point in the truthful data figure
Predictablity rate.
Preferably, the prediction model prediction result according to the error amount and the predictablity rate Calculation Estimation
Evaluation of estimate the step of include:
Subtract the error amount with one and obtain the corresponding error accuracy rate of the error amount, obtain the error accuracy rate with
The corresponding weight of the predictablity rate；
The error accuracy rate is obtained evaluating the prediction with the predictablity rate multiplied by being added after corresponding weight
The evaluation of estimate of model prediction result.
Preferably, the step of error amount calculated between the true value and the predicted value includes:
It is missed using the corresponding root mean square calculated between the true value and the predicted value of root-mean-square error calculation formula
Difference, and/or averagely missing relatively between the true value and the predicted value is calculated using average relative error calculation formula
Difference；
It is determined between the true value and the predicted value according to the root-mean-square error and/or the average relative error
Error amount；
The root-mean-square error calculation formula are as follows:
The average relative error calculation formula are as follows:
Wherein,Indicate predicted value, yiIndicate that true value, n indicate the true value or predicted value for calculating the error amount
Total number, RMSE indicate root-mean-square error, MAPE indicate average relative error.
Preferably, the prediction model prediction result according to the error amount and the predictablity rate Calculation Estimation
Evaluation of estimate the step of after, further includes:
After getting data to be predicted, according to institute's evaluation values in the prediction model selection target prediction model pair
The data to be predicted are predicted, prediction result is obtained.
Preferably, described to get data to be predicted when use accuracy rate calculates the evaluation of estimate of the prediction model
Afterwards, according to institute's evaluation values, selection target prediction model predicts the data to be predicted in the prediction model, obtains
Include: to the step of prediction result
After getting data to be predicted, select the maximum prediction model of evaluation of estimate as target in the prediction model
Prediction model；
The data to be predicted are inputted in the target prediction model, to obtain prediction result.
In addition, to achieve the above object, the present invention also provides a kind of evaluation value calculation apparatus of prediction model, the predictions
The evaluation value calculation apparatus of model includes:
Determining module, the true value concentrated for after getting training dataset, determining the training data, by pre-
At least one prediction model deposited obtains predicted value corresponding with the true value；
Computing module, for calculating the error amount between the true value and the predicted value；
Drafting module, for drawing the corresponding truthful data figure of the true value；
The computing module is also used to calculate the predictablity rate of focus in the truthful data figure；According to the mistake
The evaluation of estimate of prediction model prediction result described in difference and the predictablity rate Calculation Estimation.
In addition, to achieve the above object, the present invention also provides a kind of evaluations of estimate of prediction model to calculate equipment, the prediction
The evaluation of estimate of model calculates equipment and includes memory, processor and be stored on the memory and can transport on the processor
The evaluation of estimate calculation procedure of capable prediction model, it is real when the evaluation of estimate calculation procedure of the prediction model is executed by the processor
Now the step of evaluation value calculation method of prediction model as described above.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer readable storage medium
The evaluation of estimate calculation procedure of prediction model is stored on storage medium, the evaluation of estimate calculation procedure of the prediction model is by processor
The step of evaluation value calculation method of prediction model as described above is realized when execution.
The present invention by the prediction model prestored by obtaining concentrating with training data true after getting training dataset
The corresponding predicted value of real value calculates the error amount between true value and predicted value, the corresponding truthful data figure of construction true value, meter
The predictablity rate for calculating focus in truthful data figure, corresponds to each prediction mould of Calculation Estimation according to error amount and predictablity rate
The evaluation of estimate of type prediction result improves the evaluation of prediction model using multiple evaluation index comprehensive automation valuation prediction models
Efficiency and evaluation accuracy.
Detailed description of the invention
Fig. 1 is the flow diagram of the evaluation value calculation method first embodiment of prediction model of the present invention；
Fig. 2 is a kind of schematic diagram of medium wave of embodiment of the present invention peak dot and trough point；
Fig. 3 is the flow diagram of the evaluation value calculation method fourth embodiment of prediction model of the present invention；
Fig. 4 is the functional schematic module map of the evaluation value calculation apparatus preferred embodiment of prediction model of the present invention；
Fig. 5 is the structural schematic diagram for the hardware running environment that the embodiment of the present invention is related to.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The present invention provides a kind of evaluation value calculation method of prediction model, and referring to Fig.1, Fig. 1 is prediction model of the present invention
The flow diagram of evaluation value calculation method first embodiment.
The embodiment of the invention provides the embodiments of the evaluation value calculation method of prediction model, it should be noted that although
Logical order is shown in flow charts, but in some cases, can be executed with the sequence for being different from herein it is shown or
The step of description.
The evaluation value calculation method of prediction model is applied in the evaluation of estimate computing platform of prediction model, and prediction model is commented
Value calculation platform can be in server or terminal, terminal may include such as mobile phone, tablet computer, laptop, the palm
The mobile terminals such as upper computer, personal digital assistant (Personal Digital Assistant, PDA), and such as number TV,
The fixed terminals such as desktop computer.In each embodiment of the evaluation value calculation method of prediction model, for ease of description, save
Slightly executing subject is illustrated each embodiment.The evaluation value calculation method of prediction model includes:
Step S10 determines the true value that the training data is concentrated, passes through what is prestored after getting training dataset
At least one prediction model obtains predicted value corresponding with the true value, and calculates between the true value and the predicted value
Error amount.
After getting training dataset, the true value that training data is concentrated is determined, pass through at least one prediction prestored
Model obtains predicted value corresponding with the true value.Wherein, training dataset can be the various data that can be used for predicting, such as weather
Data, influenza data and air quality data etc..During obtaining training dataset, it can be obtained from corresponding system, such as
Weather data is obtained from meteorological office system, influenza data light signal is obtained from hospital system.The quantity of training data concentration data
It is arranged according to specific needs, such as may be configured as training data concentration and contain 100 data, 150 data or 500 numbers
According to etc..It is understood that the data that acquired training data is concentrated all are true numerical value, it is as described in this embodiment
True value is a part that training data is concentrated, which is true value corresponding with predicted value.True value can be according to specific
It needs to concentrate in training data and choose, it, such as can last 10 numbers of training data concentration to determine the true value of training data concentration
According to, last 16 data or last 20 data as true value.Prediction model be it is pre-stored, by prediction model,
The data cases of the data prediction future time section of previous time period can be used.Prediction model can be deep learning model or machine
Device learning model.Machine learning model includes but is not limited to support vector machines (SVM, Support Vector Machine), Piao
Plain Bayes (NB, Naive Bayesian), k nearest neighbour classification algorithm (KNN, k-NearestNeighbor), decision tree (DT,
Decision Tree), integrated model (RF (Random Forest, random forest)/GDBT (Gradient Boosting
Decision Tree) etc.), deep learning model includes but is not limited to convolutional neural networks (CNN, Convolutional
Neural Network), Recognition with Recurrent Neural Network (Recurrent Neural Networks) and recurrent neural network
(Recursive Neural Networks)。
After in the present embodiment using 100 days weather datas as training dataset, using first 80 days day destinys
According to, in input prediction model, predict 20 days weather datas next, then predict come 20 days weather datas for prediction
Value, 20 days weather datas are corresponding true value below in this 100 days.It is understood that at this point, the input of prediction model
It is the preceding 80 days weather datas of training data concentration of weather data, the output of prediction model is prediction resulting 20 days below
The output result of weather data, prediction model is predicted value.It follows that predicted value and true value are to belong to same time dimension
Degree, predicted value output is exported from front to back according to the sequencing of time, the predicted value of such as first output be this 100
The 81st corresponding predicted value of weather data in a data, the predicted value of second output are the 82nd day in this 100 data
Destiny is according to corresponding predicted value.
After obtaining predicted value and true value, the error amount between true value and predicted value is calculated.Specifically, it can be used flat
Equal absolute error calculation formula calculates the error amount between true value and predicted value, mean absolute error calculation formula are as follows:Wherein,Indicate predicted value, yiIndicate true value, n indicate for calculate error amount true value or
The total number of predicted value, MAE are to calculate resulting mean absolute error.It should be noted that in the present embodiment, can will be averaged
Absolute error is as the error amount between true value and predicted value.It is understood that predicted value and the number of true value are phases
Deng, i.e. the corresponding true value of a predicted value.
Step S20 draws the corresponding truthful data figure of the true value, and calculates focus in the truthful data figure
Predictablity rate.
After determining the true value that training data is concentrated, the corresponding datagram of true value is drawn, and the datagram is denoted as
Truthful data figure calculates the predictablity rate of focus in truthful data figure.Drawing the corresponding truthful data figure mistake of true value
Cheng Zhong, using the corresponding numerical value of true value as the longitudinal axis, to collect the corresponding acquisition time of the true value as horizontal axis.Wherein,
Focus is arranged according to specific needs, such as can by truthful data figure wave crest point, trough point and/or turn to select and be a little set as
Focus.It should be noted that in the present embodiment, turning point includes wave crest point and trough point.Wave crest point and trough point show
It is intended to as shown in Fig. 2, as shown in Figure 2, wave crest point is the point of truthful data figure protrusions, trough point is that truthful data figure is dented
Point.
Further, the corresponding prediction data figure of predicted value can also be drawn, in order to pay close attention in determining truthful data figure
After the corresponding true value of point, the corresponding predicted value of the focus is quickly determined by prediction data figure.
Further, when the focus is the wave crest point in the truthful data figure, the truthful data figure is calculated
The step of predictablity rate of middle focus includes:
Step a obtains the corresponding wave crest predicted value of the truthful data figure medium wave peak dot, when the wave crest point corresponds to previous
Between the first true value and the corresponding wave crest true value of the wave crest point in the period.
Specifically, when focus is the wave crest point in truthful data figure, it is corresponding to obtain truthful data figure medium wave peak dot
Wave crest predicted value, wave crest point correspond to the first true value and the corresponding wave crest true value of wave crest point in previous time period.Such as 20
There are 5 wave crest points in the truthful data figure of the true value construction of a weather data, if first wave crest point is this 20 weather
The 3rd true value in data, then this 3rd true value is wave crest true value, and corresponding wave crest predicted value is in predicted value
3rd.The length of time cycle is arranged according to specific needs, if the time cycle may be configured as 7 days a cycles, 10 days one
A period or 15 days a cycles etc..The 2nd in the 5th time cycle is such as concentrated for training data when first wave crest point
When numerical value, the first true value in the corresponding previous time period of wave crest point is the 2nd numerical value in the 4th time cycle.
Step b calculates the first difference between the wave crest predicted value and first true value, and calculates the wave
The second difference between peak true value and first true value.
After getting wave crest predicted value, the first true value and wave crest true value, it is true that wave crest predicted value is subtracted first
Value is denoted as the first difference to calculate the difference between wave crest predicted value and the first true value, and wave crest true value is subtracted the
One true value is denoted as the second difference to calculate the difference between wave crest true value and the first true value.
Step c, the product for calculating first difference and second difference are greater than zero corresponding target wave crest point quantity,
By the target wave crest point quantity divided by all wave crest point quantity in the truthful data figure, the truthful data figure medium wave is obtained
The predictablity rate of peak dot.
After the first difference and the second difference is calculated, product between the first difference and the second difference is calculated, determination multiplies
Product is greater than zero product quantity.In embodiments of the present invention, product is greater than zero corresponding wave crest point and is denoted as target wave crest point, because
This, product quantity of the product greater than zero is target wave crest point quantity.After obtaining target wave crest point quantity, by target wave crest point
Quantity is divided by wave crest point quantity all in truthful data figure, to obtain the predictablity rate of truthful data figure medium wave peak dot.Such as work as
All wave crest point quantity are 5 in truthful data figure, when target wave crest point quantity is 2, then the prediction of truthful data figure medium wave peak dot
Accuracy rate is 2 ÷ 5=0.4.
It should be noted that calculating truthful data figure medium wave valley point when focus is the trough point in truthful data figure
Predictablity rate process with calculate the process of truthful data figure medium wave peak point prediction accuracy rate it is similar, in embodiments of the present invention
No longer it is described in detail.
Step S30, according to prediction model prediction result described in the error amount and the predictablity rate Calculation Estimation
Evaluation of estimate.
When the error amount being calculated between true value and predicted value, and focus in truthful data figure is calculated
After predictablity rate, according to the evaluation of estimate of error amount and predictablity rate Calculation Estimation prediction model prediction result.It is understood that
, when there are multiple prediction models, since each prediction model is there are corresponding error amount and predictablity rate, because
This can be according to the corresponding error amount of each prediction model and predictablity rate, the corresponding each prediction model prediction result of Calculation Estimation
Evaluation of estimate.
Further, step S30 includes:
Step d, use one subtract the error amount and obtain the corresponding error accuracy rate of the error amount, and it is quasi- to obtain the error
True rate weight corresponding with the predictablity rate.
Specifically, the evaluation of estimate that prediction model is calculated according to accuracy rate, then subtract error amount with 1 and obtain the error amount pair
The error accuracy rate answered obtains error accuracy rate and the corresponding weight of predictablity rate.Wherein, error accuracy rate and prediction are accurate
The corresponding weight of rate is arranged according to specific needs.It should be noted that predictablity rate can only be the corresponding wave of wave crest point
Peak predictablity rate can also only be the corresponding trough predictablity rate of trough point, or including wave crest predictablity rate and trough
Predictablity rate.Such as when predictablity rate is wave crest predictablity rate, error accuracy rate and wave crest predictablity rate are corresponding
Weight can be respectively set to 0.6 and 0.4；When predictablity rate includes wave crest predictablity rate and trough predictablity rate, accidentally
Poor accuracy rate, wave crest predictablity rate and trough predictablity rate respective weights may be configured as 0.5,0.3 and 0.2.
Step e obtains evaluation institute by the error accuracy rate with the predictablity rate multiplied by being added after corresponding weight
State the evaluation of estimate of prediction model prediction result.
After obtaining error accuracy rate and predictablity rate respective weights, error accuracy rate is weighed multiplied by corresponding error
Weight, obtains error ratio；And predictablity rate is obtained into predicted ratio multiplied by corresponding prediction weight；Then by error ratio
Value is added with predicted ratio, obtains the evaluation of estimate of corresponding prediction model.It follows that the evaluation of estimate of prediction model=(1- error
Value) × Error weight+predictablity rate × prediction weight.It should be noted that by be in this present embodiment using accuracy rate come
The evaluation of estimate of prediction model is calculated, therefore, when the evaluation of estimate of prediction model is bigger, illustrates the prediction data that the prediction model obtains
The accuracy rate of more closing to reality situation, gained prediction result is higher.
Further, the evaluation of estimate of error amount Calculation Estimation prediction model prediction result also can be used, at this time prediction model
Evaluation of estimate=error amount × Error weight+(1- predictablity rate) × prediction weight.At this point, the evaluation of estimate of prediction model is got over
It is low, illustrate that the accuracy rate for the prediction result that the prediction model obtains is higher.
The present embodiment by the prediction model prestored by obtaining concentrating with training data after getting training dataset
The corresponding predicted value of true value calculates the error amount between true value and predicted value, constructs the corresponding truthful data figure of true value,
The predictablity rate for calculating focus in truthful data figure, corresponds to each prediction of Calculation Estimation according to error amount and predictablity rate
The evaluation of estimate of model prediction result improves commenting for prediction model using multiple evaluation index comprehensive automation valuation prediction models
Valence efficiency and evaluation accuracy.
Further, the evaluation value calculation method second embodiment of prediction model of the present invention is proposed.
The evaluation value calculation method second embodiment of the prediction model and the evaluation value calculation method of the prediction model
The difference of first embodiment is, when the focus is the turning point in the truthful data figure, calculates the true number
According to including: the step of the predictablity rate of focus in figure
Step f, when obtaining that the corresponding turnover predicted value of turning point, the turning point correspond to previous in the truthful data figure
Between the second true value and the corresponding turnover true value of the turning point in the period.
When focus is to obtain the corresponding turnover of turning point in truthful data figure and predict in truthful data figure when turning to select
Value, the turning point correspond to true value and the corresponding turnover true value of turning point in previous time period.In the embodiment of the present invention
In, turning point is corresponded into the true value in previous time period and is denoted as the second true value.Turning point predicted value and the second true value
It is corresponding with the determination process of wave crest point predicted value and the first true value similar, it is no longer described in detail in the present embodiment.
Step g calculates the third difference between the turnover predicted value and second true value, and calculates described turn
Roll over the 4th difference between true value and second true value.
After getting turnover predicted value, the second true value and turnover true value, it is true that turnover predicted value is subtracted second
Value is denoted as third difference to calculate the difference between turnover predicted value and the second true value；And turnover true value is subtracted
Second true value is denoted as the 4th difference to calculate the difference between turnover true value and the second true value.
Step h, the product for calculating the third difference and the 4th difference are greater than zero corresponding target inflection point quantity,
By the target inflection point quantity divided by the total quantity of turning point in the truthful data figure, the truthful data figure transfer is obtained
The predictablity rate of break.
After calculating third difference and four differences, the product between third difference and the 4th difference is calculated, determination multiplies
Product is greater than zero product quantity.It is understood that product quantity of the product greater than zero is between third difference and the 4th difference
Target inflection point quantity.After determining target inflection point quantity, by target inflection point quantity divided by turning point in truthful data figure
Total quantity, obtain the predictablity rate of turning point in truthful data figure, i.e. predictablity rate=target inflection point of turning point
The total quantity of quantity ÷ turning point.
It should be noted that predictablity rate, the trough of wave crest point may be selected during calculating prediction model evaluation of estimate
One or more calculating factor as Calculation Estimation value in the predictablity rate of point and the predictablity rate of turning point.Such as
When the predictablity rate and trough that select wave crest point predictablity rate as Calculation Estimation value calculating because of the period of the day from 11 p.m. to 1 a.m, preset
The wave crest weight of good wave crest point predictablity rate and the trough weight of trough point prediction accuracy rate, it is to be understood that wave crest power
The sum of weight and trough weight are equal to 1, at this point, the evaluation of estimate of prediction model=(1- error amount) × Error weight+(wave crest point it is pre-
Survey accuracy rate × wave crest weight+trough point predictablity rate × trough weight) × prediction weight.
The present embodiment, as focus, is calculated and is transferred in truthful data figure by using the turning point in truthful data figure
The predictablity rate of point, calculates prediction mould by the error amount between the predictablity rate and true value and predicted value of turning point
The evaluation of estimate of type provides the method that another calculates prediction model evaluation of estimate and selects for user relative to embodiment one, real
The diversity of prediction model evaluation value calculation method is showed.
Further, the evaluation value calculation method 3rd embodiment of prediction model of the present invention is proposed.
The evaluation value calculation method 3rd embodiment of the prediction model and the evaluation value calculation method of the prediction model
The step of difference of first or second embodiments is, calculates the error amount between the true value and the predicted value include:
Step i is calculated equal between the true value and the predicted value using root-mean-square error calculation formula correspondence
Square error, and/or being averaged between the true value and the predicted value is calculated using average relative error calculation formula
Relative error.
During calculating the error amount between true value and predicted value, root-mean-square error calculation formula can be used and calculate
Root-mean-square error between true value and predicted value, and/or true value and pre- is calculated using average relative error calculation formula
Average relative error between measured value.Wherein, root-mean-square error calculation formula are as follows:It is average opposite
Error calculation formula are as follows:
In the opposite error calculation formula of root-mean-square error calculation formula peace,Indicate predicted value, yiIndicate true
Value, n indicate that, for calculating the true value of error amount or the total number of predicted value, RMSE indicates root-mean-square error, and MAPE indicates flat
Equal relative error.
Step j determines the true value and the prediction according to the root-mean-square error and/or the average relative error
Error amount between value.
After calculating root-mean-square error and/or average relative error, root-mean-square error can be used as true value and pre-
Average relative error can also be used as the error amount between true value and predicted value in error amount between measured value；Or it can be with
Error amount between true value and predicted value is calculated using root-mean-square error and average relative error together, at this point, can set in advance
The corresponding root mean square weight of root-mean-square error has been set, and has pre-set the corresponding average relative weighting of average relative error,
Error amount=root-mean-square error × root mean square weight+average relative error × average opposite power between true value and predicted value
Weight.Wherein, root mean square weight peace relative weighting is arranged according to specific needs, in the present embodiment to root mean square weight
The size of peaceful relative weighting is not particularly limited.It is understood that in embodiments of the present invention, can be missed in average absolute
One or more error is selected to determine between true value and predicted value between difference, root-mean-square error and average relative error
Error amount.
The present embodiment by root-mean-square error calculation formula and/or average relative error calculation formula calculate true value and
Error amount between predicted value provides a variety of calculation methods to calculate the error amount between true value and predicted value, realizes
The diversity of error amount calculation between true value and predicted value.
Further, the evaluation value calculation method fourth embodiment of prediction model of the present invention is proposed.
The evaluation value calculation method fourth embodiment of the prediction model and the evaluation value calculation method of the prediction model
The difference of first, second or third embodiment is, referring to Fig. 3, the evaluation value calculation method of prediction model further include:
Step S40, after getting data to be predicted, according to institute's evaluation values, selection target is pre- in the prediction model
It surveys model to predict the data to be predicted, obtains prediction result.
After getting data to be predicted, a prediction is selected in prediction model according to the evaluation of estimate of each prediction model
Data to be predicted are inputted in target prediction model as target prediction model, are predicted, obtained with treating prediction data by model
To corresponding prediction result.Wherein, which can send for other systems, or defeated manually by counterpart staff
Enter.It is understood that the output of target prediction model is prediction result.It should be noted that according to evaluation of estimate pre-
In survey model during selection target prediction model, it can be selected according to the size of evaluation of estimate.It is such as each when being calculated using error amount
When the corresponding evaluation of estimate of a prediction model, it may be configured as selecting the smallest prediction model of evaluation of estimate as target in prediction model
Prediction model.Further, it may be alternatively provided as that evaluation of estimate is selected to come small prediction model second from the bottom as target prediction mould
Type.
Further, when calculating the evaluation of estimate of the prediction model using accuracy rate, step S40 includes:
Step k selects the maximum prediction model of evaluation of estimate to make after getting data to be predicted in the prediction model
For target prediction model.
When calculating the evaluation of estimate of prediction model using accuracy rate, after getting data to be predicted, in prediction model
Select the maximum prediction model of evaluation of estimate as target prediction model.It is understood that the maximum prediction mould of evaluation of estimate at this time
Type is the highest model of prediction result accuracy rate.The evaluation of estimate of two or more prediction model is equal if it exists, and
The evaluation of estimate is maximum evaluation of estimate, can randomly select a conduct in the two or more than two prediction models at this time
Target prediction model.
Step l inputs the data to be predicted in the target prediction model, to obtain prediction result.
After determining target prediction model, data to be predicted are inputted in target prediction model, to obtain prediction result.
The present embodiment treats prediction data progress by selecting target prediction model in prediction model according to evaluation of estimate
Prediction, obtains prediction result, realizes and selects the highest prediction model of prediction result accuracy rate to number to be predicted according to evaluation of estimate
According to being predicted, the accuracy rate of gained prediction result is improved.
In addition, referring to Fig. 4, the present invention also provides a kind of evaluation value calculation apparatus of prediction model, the prediction model
Evaluating value calculation apparatus includes:
Determining module 10, the true value concentrated for after getting training dataset, determining the training data, passes through
At least one prediction model prestored obtains predicted value corresponding with the true value；
Computing module 20, for calculating the error amount between the true value and the predicted value；
Drafting module 30, for drawing the corresponding truthful data figure of the true value；
The computing module 20 is also used to calculate the predictablity rate of focus in the truthful data figure；According to described
The evaluation of estimate of prediction model prediction result described in error amount and the predictablity rate Calculation Estimation.
Further, when the focus is the wave crest point in the truthful data figure, the computing module 20 includes:
First acquisition unit, for obtaining the corresponding wave crest predicted value of the truthful data figure medium wave peak dot, the wave crest
The first true value and the corresponding wave crest true value of the wave crest point in the corresponding previous time period of point；
First computing unit, for calculating the first difference between the wave crest predicted value and first true value, with
And the second difference between the calculating wave crest true value and first true value；Calculate first difference and described second
The product of difference is greater than zero corresponding target wave crest point quantity, by the target wave crest point quantity divided by the truthful data figure
All wave crest point quantity obtain the predictablity rate of the truthful data figure medium wave peak dot.
Further, when the focus is the turning point in the truthful data figure, the computing module 20 is also wrapped
It includes:
Second acquisition unit, for obtaining the corresponding turnover predicted value of turning point, the turnover in the truthful data figure
The second true value and the corresponding turnover true value of the turning point in the corresponding previous time period of point；
Second computing unit, for calculating the third difference between the turnover predicted value and second true value, with
And calculate the 4th difference transferred between true value and second true value；Calculate the third difference and the described 4th
The product of difference is greater than zero corresponding target inflection point quantity, by the target inflection point quantity divided by the truthful data figure
The total quantity of turning point obtains the predictablity rate of turning point in the truthful data figure.
Further, the computing module 20 further include:
Third computing unit obtains the corresponding error accuracy rate of the error amount for subtracting the error amount with one；
Third acquiring unit, for obtaining error accuracy rate weight corresponding with the predictablity rate；
The third computing unit is also used to the error accuracy rate and the predictablity rate multiplied by corresponding weight
After be added, obtain the evaluation of estimate for evaluating the prediction model prediction result.
Further, the computing module 20 further include:
4th computing unit, for calculating the true value and the prediction using root-mean-square error calculation formula correspondence
Root-mean-square error between value, and/or the true value and the predicted value are calculated using average relative error calculation formula
Between average relative error；
Determination unit, for according to the root-mean-square error and/or the average relative error determine the true value and
Error amount between the predicted value；
The root-mean-square error calculation formula are as follows:
The average relative error calculation formula are as follows:
Wherein,Indicate predicted value, yiIndicate that true value, n indicate the true value or predicted value for calculating the error amount
Total number, RMSE indicate root-mean-square error, MAPE indicate average relative error.
Further, the evaluation value calculation apparatus of the prediction model further include:
Selecting module, for being selected in the prediction model according to institute's evaluation values after getting data to be predicted
Target prediction model predicts the data to be predicted, obtains prediction result.
Further, the selecting module includes:
Selecting unit, for selecting evaluation of estimate maximum pre- in the prediction model after getting data to be predicted
Model is surveyed as target prediction model；
Input unit, for inputting the data to be predicted in the target prediction model, to obtain prediction result.
It should be noted that each embodiment of the evaluation value calculation apparatus of prediction model and the evaluation of above-mentioned prediction model
Each embodiment of value calculating method is essentially identical, and in this not go into detail.
In addition, the present invention also provides a kind of evaluations of estimate of prediction model to calculate equipment.As shown in figure 5, Fig. 5 is of the invention real
Apply the structural schematic diagram for the hardware running environment that a scheme is related to.
It should be noted that the structure for the hardware running environment that Fig. 5 can calculate equipment for the evaluation of estimate of prediction model is shown
It is intended to.The evaluation of estimate of prediction model of the embodiment of the present invention, which calculates equipment, can be PC, the terminal devices such as portable computer.
As shown in figure 5, it may include: processor 1001, such as CPU, storage that the evaluation of estimate of the prediction model, which calculates equipment,
Device 1005, user interface 1003, network interface 1004, communication bus 1002.Wherein, communication bus 1002 is for realizing these groups
Connection communication between part.User interface 1003 may include display screen (Display), input unit such as keyboard
(Keyboard), optional user interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 is optional
May include standard wireline interface and wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory,
It is also possible to stable memory (non-volatile memory), such as magnetic disk storage.Memory 1005 optionally may be used also
To be independently of the storage device of aforementioned processor 1001.
Optionally, the evaluation of estimate calculating equipment of prediction model can also include camera, (Radio Frequency, is penetrated RF
Frequently circuit), sensor, voicefrequency circuit, WiFi module etc..
It will be understood by those skilled in the art that the evaluation of estimate of prediction model shown in Fig. 5 calculates device structure not structure
The evaluation of estimate of pairs of prediction model calculates the restriction of equipment, may include components more more or fewer than diagram, or combine certain
A little components or different component layouts.
As shown in figure 5, as may include that operating system, network are logical in a kind of memory 1005 of computer storage medium
Believe the evaluation of estimate calculation procedure of module, Subscriber Interface Module SIM and prediction model.Wherein, operating system is to manage and control prediction
The evaluation of estimate of model calculates the program of device hardware and software resource, supports the evaluation of estimate calculation procedure of prediction model and other
The operation of software or program.
The evaluation of estimate of prediction model shown in Fig. 5 calculates in equipment, and user interface 1003 can be used for connecting training data
Collect correspondence system, obtains training dataset from training dataset correspondence system；Network interface 1004 is mainly used for connection backstage
Server carries out data communication with background server；Processor 1001 can be used for calling the prediction stored in memory 1005
The evaluation of estimate calculation procedure of model, and the step of executing the evaluation value calculation method of prediction model as described above.
The evaluation of estimate of prediction model of the present invention calculates equipment specific embodiment and the evaluation of estimate of above-mentioned prediction model calculates
Each embodiment of method is essentially identical, and details are not described herein.
In addition, the embodiment of the present invention also proposes a kind of computer readable storage medium, the computer readable storage medium
On be stored with the evaluation of estimate calculation procedure of prediction model, it is real when the evaluation of estimate calculation procedure of the prediction model is executed by processor
Now the step of evaluation value calculation method of prediction model as described above.
Computer readable storage medium specific embodiment of the present invention and the evaluation value calculation method of above-mentioned prediction model are each
Embodiment is essentially identical, and details are not described herein.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal device (can be mobile phone, computer, clothes
Business device, air conditioner or the network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of evaluation value calculation method of prediction model, which is characterized in that the evaluation value calculation method packet of the prediction model
Include following steps:
After getting training dataset, the true value that the training data is concentrated is determined, pass through at least one prediction prestored
Model obtains predicted value corresponding with the true value, and calculates the error amount between the true value and the predicted value；
The corresponding truthful data figure of the true value is drawn, and calculates the predictablity rate of focus in the truthful data figure；
According to the evaluation of estimate of prediction model prediction result described in the error amount and the predictablity rate Calculation Estimation.
2. the evaluation value calculation method of prediction model as described in claim 1, which is characterized in that when the focus is described
When wave crest point in truthful data figure, described the step of calculating the predictablity rate of focus in the truthful data figure, includes:
Obtain the corresponding wave crest predicted value of the truthful data figure medium wave peak dot, the wave crest point corresponds in previous time period
First true value and the corresponding wave crest true value of the wave crest point；
Calculate the first difference between the wave crest predicted value and first true value, and calculate the wave crest true value and
The second difference between first true value；
The product for calculating first difference and second difference is greater than zero corresponding target wave crest point quantity, by the target
Wave crest point quantity obtains the prediction of the truthful data figure medium wave peak dot divided by all wave crest point quantity in the truthful data figure
Accuracy rate.
3. the evaluation value calculation method of prediction model as described in claim 1, which is characterized in that when the focus is described
When turning point in truthful data figure, described the step of calculating the predictablity rate of focus in the truthful data figure, includes:
The corresponding turnover predicted value of turning point, the turning point in the truthful data figure is obtained to correspond in previous time period
Second true value and the corresponding turnover true value of the turning point；
Calculate the third difference between the turnover predicted value and second true value, and calculate the turnover true value with
The 4th difference between second true value；
The product for calculating the third difference and the 4th difference is greater than zero corresponding target inflection point quantity, by the target
Turning point quantity obtains the prediction of turning point in the truthful data figure divided by the total quantity of turning point in the truthful data figure
Accuracy rate.
4. the evaluation value calculation method of prediction model as described in claim 1, which is characterized in that described according to the error amount
Include: with the step of evaluation of estimate of prediction model prediction result described in the predictablity rate Calculation Estimation
Subtract the error amount with one and obtain the corresponding error accuracy rate of the error amount, obtain the error accuracy rate with it is described
The corresponding weight of predictablity rate；
The error accuracy rate is obtained evaluating the prediction model with the predictablity rate multiplied by being added after corresponding weight
The evaluation of estimate of prediction result.
5. the evaluation value calculation method of prediction model as described in claim 1, which is characterized in that described to calculate the true value
The step of error amount between the predicted value includes:
The root-mean-square error calculated between the true value and the predicted value is corresponded to using root-mean-square error calculation formula,
And/or the average relative error between the true value and the predicted value is calculated using average relative error calculation formula；
The mistake between the true value and the predicted value is determined according to the root-mean-square error and/or the average relative error
Difference；
The root-mean-square error calculation formula are as follows:
The average relative error calculation formula are as follows:
Wherein,Indicate predicted value, yiIndicate true value, n indicate for calculate the error amount true value or predicted value it is total
Number, RMSE indicate root-mean-square error, and MAPE indicates average relative error.
6. such as the evaluation value calculation method of prediction model described in any one of claim 1 to 5, which is characterized in that the basis
After the step of evaluation of estimate of prediction model prediction result described in the error amount and the predictablity rate Calculation Estimation, also wrap
It includes:
After getting data to be predicted, according to institute's evaluation values in the prediction model selection target prediction model to described
Data to be predicted are predicted, prediction result is obtained.
7. the evaluation value calculation method of prediction model as claimed in claim 6, which is characterized in that when using accuracy rate calculating institute
It is described after getting data to be predicted when stating the evaluation of estimate of prediction model, according to institute's evaluation values in the prediction model
The step of selection target prediction model is predicted to the data to be predicted, obtains prediction result include:
After getting data to be predicted, select the maximum prediction model of evaluation of estimate as target prediction in the prediction model
Model；
The data to be predicted are inputted in the target prediction model, to obtain prediction result.
8. a kind of evaluation value calculation apparatus of prediction model, which is characterized in that the evaluation value calculation apparatus packet of the prediction model
It includes:
Determining module, the true value concentrated for after getting training dataset, determining the training data, passes through what is prestored
At least one prediction model obtains predicted value corresponding with the true value；
Computing module, for calculating the error amount between the true value and the predicted value；
Drafting module, for drawing the corresponding truthful data figure of the true value；
The computing module is also used to calculate the predictablity rate of focus in the truthful data figure；According to the error amount
With the evaluation of estimate of prediction model prediction result described in the predictablity rate Calculation Estimation.
9. a kind of evaluation of estimate of prediction model calculates equipment, which is characterized in that the evaluation of estimate of the prediction model calculates equipment packet
Include memory, processor and the evaluation of estimate meter for being stored in the prediction model that can be run on the memory and on the processor
Program is calculated, is realized when the evaluation of estimate calculation procedure of the prediction model is executed by the processor as any in claim 1 to 7
The step of evaluation value calculation method of prediction model described in.
10. a kind of computer readable storage medium, which is characterized in that be stored with prediction mould on the computer readable storage medium
The evaluation of estimate calculation procedure of type realizes such as claim 1 when the evaluation of estimate calculation procedure of the prediction model is executed by processor
To prediction model described in any one of 7 evaluation value calculation method the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811485623.7A CN109816196A (en) | 2018-12-04 | 2018-12-04 | Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811485623.7A CN109816196A (en) | 2018-12-04 | 2018-12-04 | Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109816196A true CN109816196A (en) | 2019-05-28 |
Family
ID=66601956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811485623.7A Pending CN109816196A (en) | 2018-12-04 | 2018-12-04 | Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109816196A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110362724A (en) * | 2019-07-23 | 2019-10-22 | 国家卫星海洋应用中心 | A kind of data filtering method, device, electronic equipment and readable storage medium storing program for executing |
| CN110390425A (en) * | 2019-06-20 | 2019-10-29 | 阿里巴巴集团控股有限公司 | Prediction technique and device |
| CN110472191A (en) * | 2019-07-02 | 2019-11-19 | 北京大学 | A kind of the service evaluation calculation method and device of dynamic self-adapting |
| CN111259302A (en) * | 2020-01-19 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Information pushing method and device and electronic equipment |
| CN111482467A (en) * | 2020-04-15 | 2020-08-04 | 中冶赛迪重庆信息技术有限公司 | Rolling mill parameter management method |
| CN111917121A (en) * | 2020-06-30 | 2020-11-10 | 嘉善数能技术有限公司 | Control method for power consumption time period peak clipping and valley filling power supply of communication base station |
| WO2022057427A1 (en) * | 2020-09-15 | 2022-03-24 | 华为技术有限公司 | Method, apparatus and device for power prediction |
| CN114358445A (en) * | 2022-03-21 | 2022-04-15 | 山东建筑大学 | Business process residual time prediction model recommendation method and system |
| CN114818993A (en) * | 2022-06-24 | 2022-07-29 | 启东通奕自动化设备有限公司 | Service life analysis method for hammer mill |
-
2018
- 2018-12-04 CN CN201811485623.7A patent/CN109816196A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390425A (en) * | 2019-06-20 | 2019-10-29 | 阿里巴巴集团控股有限公司 | Prediction technique and device |
| CN110472191A (en) * | 2019-07-02 | 2019-11-19 | 北京大学 | A kind of the service evaluation calculation method and device of dynamic self-adapting |
| CN110472191B (en) * | 2019-07-02 | 2021-03-12 | 北京大学 | Dynamic self-adaptive service evaluation calculation method and device |
| CN110362724A (en) * | 2019-07-23 | 2019-10-22 | 国家卫星海洋应用中心 | A kind of data filtering method, device, electronic equipment and readable storage medium storing program for executing |
| CN111259302A (en) * | 2020-01-19 | 2020-06-09 | 腾讯科技（深圳）有限公司 | Information pushing method and device and electronic equipment |
| CN111482467A (en) * | 2020-04-15 | 2020-08-04 | 中冶赛迪重庆信息技术有限公司 | Rolling mill parameter management method |
| CN111917121A (en) * | 2020-06-30 | 2020-11-10 | 嘉善数能技术有限公司 | Control method for power consumption time period peak clipping and valley filling power supply of communication base station |
| CN111917121B (en) * | 2020-06-30 | 2023-08-18 | 嘉善利天万世新能源有限公司 | Power supply control method for peak clipping and valley filling in power utilization period of communication base station |
| WO2022057427A1 (en) * | 2020-09-15 | 2022-03-24 | 华为技术有限公司 | Method, apparatus and device for power prediction |
| CN114358445A (en) * | 2022-03-21 | 2022-04-15 | 山东建筑大学 | Business process residual time prediction model recommendation method and system |
| CN114818993A (en) * | 2022-06-24 | 2022-07-29 | 启东通奕自动化设备有限公司 | Service life analysis method for hammer mill |
| CN114818993B (en) * | 2022-06-24 | 2022-09-02 | 启东通奕自动化设备有限公司 | Service life analysis method for hammer mill |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109816196A (en) | Evaluation value calculation method, device, equipment and the readable storage medium storing program for executing of prediction model | |
| Jia et al. | Ant colony optimization algorithm for scheduling jobs with fuzzy processing time on parallel batch machines with different capacities | |
| Özcan et al. | A genetic algorithm for the stochastic mixed-model U-line balancing and sequencing problem | |
| Liu et al. | Reinforcement learning-based cell selection in sparse mobile crowdsensing | |
| Pang et al. | Evaluation of the results of multi-attribute group decision-making with linguistic information | |
| CN104915735B (en) | The early warning analysis method of enterprise operation analyzing and alarming system | |
| Li et al. | Multi-criteria group decision making based on trapezoidal intuitionistic fuzzy information | |
| CN111259222B (en) | Article recommendation method, system, electronic equipment and storage medium | |
| US11830033B2 (en) | Theme recommendation method and apparatus | |
| CN109615020A (en) | Characteristic analysis method, device, equipment and medium based on machine learning model | |
| García-Pedrajas et al. | Multi-objective cooperative coevolution of artificial neural networks (multi-objective cooperative networks). | |
| Ju et al. | Projection method for multiple criteria group decision making with incomplete weight information in linguistic setting | |
| CN113191828B (en) | User electricity price value grade label construction method, device, equipment and medium | |
| CN110413510B (en) | Data processing method, device and equipment | |
| CN110069715A (en) | A kind of method of information recommendation model training, the method and device of information recommendation | |
| CN107273979A (en) | The method and system of machine learning prediction are performed based on service class | |
| Tang et al. | Decision making with multiplicative hesitant fuzzy linguistic preference relations | |
| CN114095381B (en) | Multitask model training method, multitask prediction method and related products | |
| CN110020877A (en) | The prediction technique of clicking rate, the determination method of clicking rate and server | |
| CN109379410B (en) | Information-pushing method, device, server and storage medium | |
| CN112686371A (en) | Network structure search method, device, equipment, storage medium and program product | |
| Nagode et al. | The rebmix algorithm for the multivariate finite mixture estimation | |
| CN110866625A (en) | Promotion index information generation method and device | |
| Saber et al. | Optimization of economic load dispatch of higher order general cost polynomials and its sensitivity using modified particle swarm optimization | |
| Lang et al. | Improved neural networks with random weights for short-term load forecasting |
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 |