Wind power prediction method and system considering climbing feature recognition and error correction

Abstract

The invention provides a wind power prediction method and system considering climbing feature recognition and error correction. The method comprises the following steps: step 1, carrying out climbing recognition on a climbing event of a historical wind power sequence according to features; based on the identification result, classifying all the wind power sequences into an upward climbing stage, a downward climbing stage and a non-climbing stage; step 2, acquiring meteorological factors of wind power; 3, respectively establishing prediction models for the uphill stage, the downhill stage and the non-uphill stage, and respectively inputting the identified uphill stage, downhill stage and non-uphill stage and meteorological factors into the corresponding prediction models for prediction to obtain three prediction results; and 4, correcting the three prediction results by using an error correction model to obtain a final wind power prediction result. According to the method, the wind power under the wind power climbing event can be accurately predicted, and the overall precision of short-term wind power prediction is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power prediction, and in particular relates to a wind power prediction method and system considering slope characteristic recognition and error correction. Background Art

[0002] When wind power changes rapidly, traditional forecasting methods often fail to accurately capture such changes in a timely manner, resulting in large forecasting errors. Such errors will affect the dispatching decisions of the power grid and increase the risk of power grid operation. At the same time, when wind power changes rapidly, accurate forecasting can help reduce the demand for backup capacity of the power system, reduce the wind curtailment rate, and improve the overall energy efficiency, thereby improving economic benefits. Summary of the invention

[0003] In order to overcome the technical problem that the above-mentioned prior art cannot timely and accurately capture the slope change, resulting in large prediction errors, in the first aspect, the present invention proposes a wind power prediction method considering slope feature recognition and error correction, including:

[0004] Step 1: Perform climbing identification on the climbing events of the historical wind power sequence according to the characteristics; based on the identification result, classify all wind power sequences into the climbing stage, the descending stage and the non-climbing stage;

[0005] Step 2: Obtain meteorological factors of wind power;

[0006] Step 3: Establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction, and obtain three prediction results;

[0007] Step 4: Use the error correction model to correct the three prediction results to obtain the final wind power prediction result.

[0008] Preferably, in step 3, prediction models are established for the uphill climbing stage, the downhill climbing stage and the non-climbing stage respectively, including:

[0009] The SSA-BiLSTM neural network prediction model is used to predict the climbing stage in the wind power sequence, the CNN-BiLSTM neural network prediction model is used to predict the downclimbing stage in the wind power sequence, and the LSTM neural network prediction model is used to predict the non-climbing stage in the wind power sequence.

[0010] Preferably, in step 1, the features include 4, which are amplitude change ΔP r , duration Δt, climbing rate ΔP r / Δt, climbing direction.

[0011] Preferably, in step 1, based on the identification result, classifying all wind power sequences into an up-climbing phase, a down-climbing phase and a non-climbing phase includes:

[0012] The plurality of wind power sequences are compressed according to the parallelogram rule; the up-climbing stage, the down-climbing stage and the non-climbing stage are used as constraints of the revolving door algorithm, and parallelograms are continuously constructed so that each of the wind power sequences is contained in the parallelogram; if there are points corresponding to the wind power sequences that are not contained in the parallelogram, then all points before the points corresponding to the currently calculated wind power sequence need to be compressed, and then the long wind power sequence corresponding to the compressed points is divided into an up-climbing sequence, a down-climbing sequence and a non-climbing sequence.

[0013] Preferably, after obtaining the meteorological factors of wind power, the method further includes:

[0014] Based on the correlation coefficient, a Pearson correlation coefficient algorithm or a Spearman correlation coefficient algorithm is used to calculate the correlation coefficient between each meteorological factor in each time period and the uphill climbing stage, the downhill climbing stage or the non-climbing stage;

[0015] Based on the correlation coefficient, meteorological factors having a strong correlation with the uphill climbing stage, the downhill climbing stage or the non-climbing stage are screened out.

[0016] Preferably, based on the correlation coefficient, meteorological factors having strong correlation with the uphill climbing stage, the downhill climbing stage or the non-climbing stage are screened out, including screening meteorological factors according to the principle that the closer the correlation coefficient is to 1, the stronger the correlation.

[0017] Preferably, the step 4: using an error correction model to correct the three prediction results to obtain a final wind power prediction result includes:

[0018] For the prediction results corresponding to the up-climbing stage and the down-climbing stage, the power prediction is performed using the neural network model of the sparrow algorithm and the BiLSTM neural network prediction model;

[0019] The calculation is as follows:

[0020]

[0021] Where: is the forward output of the hidden layer in the up-climbing stage and the down-climbing stage, is the backward output of the hidden layer in the up-climbing stage and the down-climbing stage; L1 indicates forward calculation, and L2 indicates backward calculation. tis the predicted value of the up-climbing stage and the down-climbing stage obtained at time t, c t is the bias optimization parameter at time t; a t represents the forward output weight; b t Represents the backward output weight; i t represents the input of the model at time t; B t ' represents the wind power prediction result; n is the number of calculations.

[0022] Preferably, after the step 4: correcting the three prediction results by using the error correction model to obtain the final wind power prediction result, the method further includes:

[0023] Construct a prediction error sequence based on the wind power prediction value of the above integrated prediction model;

[0024] Predict the prediction error sequence to obtain the error sequence prediction value;

[0025] The error prediction value is combined with the wind power prediction value to obtain the final wind power prediction result after the difference is correct.

[0026] Preferably, the error correction satisfies the following formula:

[0027] e=B t -B t '

[0028] e'(t)=f[e(tT),e(t-2T),…,e(t-nT)]

[0029] B t =B t '+e'

[0030] Where: B t is the actual value of wind power; B t ' is the preliminary forecast value of wind power; e is the calculated forecast error sequence; e(t-nT) is the nth historical forecast error before time t; e'(t) is the constructed error sequence; B t ” is the final wind power prediction result after error correction.

[0031] On the other hand, the present invention also proposes a wind power prediction system considering ramp feature recognition and error correction, comprising:

[0032] A ramp classification module is used to identify the ramp events of the historical wind power sequence according to the characteristics; based on the identification results, all wind power sequences are classified into an up-climbing stage, a down-climbing stage and a non-climbing stage;

[0033] Meteorological factor acquisition module, used to obtain meteorological factors of wind power;

[0034] A prediction result determination module is used to establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, and input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction to obtain three prediction results;

[0035] The final wind power prediction result determination module is used to correct the three prediction results using the error correction model to obtain the final wind power prediction result.

[0036] Compared with the closest prior art, the present invention has the following beneficial effects:

[0037] Being able to accurately predict wind power forecasts under wind power ramping events and actively respond to power mutations caused by ramping events can promote consumption, ensure reliable system operation, and greatly improve the overall accuracy of short-term wind power forecasts. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the overall flow chart of the present invention;

[0039] Figure 2 is a schematic diagram of the climbing characteristics of the present invention;

[0040] Figure 3 is a flow chart of the method of the present invention;

[0041] Figure 4 It is a CNN-BiLSTM model diagram of the present invention. DETAILED DESCRIPTION

[0042] The following is combined with Figure 1-4 The specific implementation methods of the present invention are further described in detail.

[0043] Embodiment 1:

[0044] The present invention proposes a wind power prediction method considering ramp feature recognition and error correction, comprising:

[0045] Step 1: Perform climbing identification on the climbing events of the historical wind power sequence according to the characteristics; based on the identification result, classify all wind power sequences into the climbing stage, the descending stage and the non-climbing stage;

[0046] Step 2: Obtain meteorological factors of wind power;

[0047] Step 3: Establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction, and obtain three prediction results;

[0048] Step 4: Use the error correction model to correct the three prediction results to obtain the final wind power prediction result.

[0049] In the above step 1, the features include 4, which are amplitude change ΔP r , duration Δt, climbing rate ΔP r / Δt, climbing direction.

[0050] 1) Duration Δt is the time when wind power fluctuates greatly;

[0051] 2) Climbing rate ΔP r / Δt is the rate of change of wind power in a certain period of time;

[0052] 3) Amplitude change value ΔP r is the change value of wind power in a certain period of time;

[0053] 4) Climbing direction: uphill climbing and downhill climbing.

[0054] The following two formulas are combined to define ramping. As long as one of the power mutation amplitude threshold or the ramp rate threshold is met, a ramping event is considered to have occurred. Due to the different ramping directions, the thresholds for the up and down ramping events are set to 20% and 15% of the rated power, respectively.

[0055] Definition 1: When the absolute value of the power difference between the start time and the end time of the time period (t, t+Δt) exceeds a given threshold, a ramp event is considered to have occurred.

[0056] P t+Δt -P t |>P threshold (1)

[0057] Where: P t is the wind power at time t, P t+Δt is the wind power at time t+Δt, P threshold is a given power threshold.

[0058] Definition 2: When the ramp rate in the time period (t, t+Δt) exceeds a given threshold, a ramp event is considered to have occurred.

[0059]

[0060] Where: R threshold is the given climbing rate threshold.

[0061] Combined with the definition of ramping, the revolving door algorithm is used to identify the ramping of the power sequence. The principle of the revolving door algorithm is to compress the wind power sequence according to the parallelogram rule. According to the above ramping definition as the constraint condition of the algorithm, a parallelogram is continuously constructed so that each wind power sample is included in this parallelogram. If there are some points that are not included in the parallelogram, the previous points need to be compressed, thereby dividing a long wind power sequence into an up-climbing sequence, a down-climbing sequence and a non-climbing sequence.

[0062] In step 1, based on the identification result, all wind power sequences are classified into an up-climbing phase, a down-climbing phase and a non-climbing phase, including:

[0063] The plurality of wind power sequences are compressed according to the parallelogram rule; the up-climbing stage, the down-climbing stage and the non-climbing stage are used as constraints of the revolving door algorithm, and parallelograms are continuously constructed so that each of the wind power sequences is contained in the parallelogram; if there are points corresponding to the wind power sequences that are not contained in the parallelogram, then all points before the points corresponding to the currently calculated wind power sequence need to be compressed, and then the long wind power sequence corresponding to the compressed points is divided into an up-climbing sequence, a down-climbing sequence and a non-climbing sequence.

[0064] In the above step 2, when performing the correlation analysis of wind power, it is first necessary to select a group of meteorological factors, which include wind speed, wind direction, temperature, air pressure, and humidity.

[0065] After obtaining the meteorological factors of wind power, the method further includes:

[0066] The correlation coefficients between each meteorological factor and different climbing stages in each time period are calculated respectively. Based on the correlation coefficients, the correlation coefficients between each meteorological factor and the climbing stage, the descending stage or the non-climbing stage in each time period are calculated using the Pearson correlation coefficient algorithm or the Spearman correlation coefficient algorithm. The range of the Pearson correlation coefficient value is generally {-1, 1}, and its formula is as follows:

[0067]

[0068] In the formula: x, y are the time series of the two selected reference variables, cov(x, y) is the covariance of the two series, x b ,y b are the means of x and y sequences respectively, a represents the time series, each event sequence corresponds to an actual value, and the specific selected meteorological factors are shown in Table 1.

[0069] By calculating and comparing the correlation between each factor and wind power in different time periods, the factors with the most significant impact in different wind power ramping stages can be identified. This segmented correlation analysis helps to reveal the driving factors of wind power changes. For example, some meteorological factors may have a stronger impact on wind power in the ramping stage, but a weaker impact in the ramping stage or non-ramp stage. Comparing the ρ calculated by each meteorological factor x,y , sorted from large to small, thus screening out the strong correlation factors of different climbing sections, and inputting the selected characteristic factors and wind power sequence into the prediction model established below. Through this analysis, it can be used to improve the wind power prediction model so that it has higher prediction accuracy under different wind power change states.

[0070] Table 1 Correlation factors

[0071] Meteorological factors Representation method 10m wind speed 10mWS 30m wind speed 30mWS 50m wind speed 50mWS 10m wind direction 10mWD 30m wind direction 30mWD 50m wind direction 50mWD temperature T Relative humidity H Atmospheric pressure P Wind WP

[0072] Based on the correlation coefficient, meteorological factors having a strong correlation with the uphill climbing stage, the downhill climbing stage or the non-climbing stage are screened out.

[0073] In the above, based on the correlation coefficient, the meteorological factors with strong correlation with the climbing stage, the downhill climbing stage or the non-climbing stage are screened out, including screening meteorological factors according to the principle that the closer the correlation coefficient is to 1, the stronger the correlation is.

[0074] In the above-mentioned step 3, prediction models are established for the up-climbing stage, the down-climbing stage and the non-climbing stage respectively, including:

[0075] The SSA-BiLSTM neural network prediction model is used to predict the climbing stage in the wind power sequence, the CNN-BiLSTM neural network prediction model is used to predict the downclimbing stage in the wind power sequence, and the LSTM neural network prediction model is used to predict the non-climbing stage in the wind power sequence.

[0076] In the above, the step 4: using the error correction model to correct the three prediction results to obtain the final wind power prediction result includes:

[0077] For the prediction results corresponding to the up-climbing stage and the down-climbing stage, the neural network model of the sparrow algorithm (SSA) and the BiLSTM neural network prediction model is used to predict the power;

[0078] Since the traditional BiLSTM neural network prediction model has insufficient accuracy in the power ramp-up stage, and the prediction accuracy depends on the selection of parameters such as the number of hidden layer units, learning rate and regularization coefficient of the BiLSTM network, these parameters are generally selected by multiple experiments, which is inefficient. Therefore, the SSA algorithm is selected to optimize the model parameters by learning the predation behavior of the sparrow population, and obtain the optimal parameters of BiLSTM to improve the prediction ability of the neural network. The specific calculation process of SSA is as follows. The sparrow population is divided into visitors, followers, observers and leavers according to their working characteristics.

[0079] Among them, the finder's location is updated as follows:

[0080]

[0081] In the formula, represents the d-th dimension position of individual i in the sparrow in the t-th iteration; a and R2 are random numbers between 0 and 1; iter max is the maximum number of iterations of the population; Q is a random number and obeys Gaussian distribution; ST is the alarm threshold and its value is between 0.5 and 1.

[0082] The joiner position is updated as follows:

[0083]

[0084] In the formula, is the worst individual sparrow position, is the best individual position of the sparrow; rand{-1,1} takes a random number between 1 and -1; n is the total number of individuals, when When the position needs to be updated; In the case of , it means that the current position is near the optimal position.

[0085] In each generation, a certain proportion of individuals will be selected as alarmers, and the alarmer position is updated as follows:

[0086]

[0087] Where β is a random number that follows a Gaussian distribution; f i is the current fitness value of a single sparrow, and f b is the best fitness value among the current sparrow individuals, f w is the worst individual fitness value; K is a random number between 1 and -1; ε is a number very close to 0 but not 0. The optimization results of the sparrow algorithm after the positions of the discoverer, joiner, and explorer are updated and brought into the BiLSTM neural network prediction model to generate a neural network model of the sparrow algorithm (SSA) and the BiLSTM neural network prediction model.

[0088] By setting multiple hidden layers between the input and output data, the convolutional neural network can automatically learn the complex nonlinear relationship between the data. Its prediction accuracy is often better than that of commonly used machine learning methods, and it can solve the problem of unsatisfactory prediction results of traditional BiLSTM during the power reduction stage.

[0089] Among them, the convolutional neural network (CNN) can improve the model training speed and reduce the model's network parameters through weight sharing and downsampling. It mainly includes convolutional layers, pooling layers, and fully connected layers.

[0090] The error of the lth layer in CNN is:

[0091] δ (l) =((W (l) ) T δ (l+1) )·f′(z (l) ) (8)

[0092] The gradient formulas of its parameters are:

[0093]

[0094] In the formula, δ (l+1) is the error of the l+1th layer, J(W,b;x,y) is the loss function, W,b,x,y are the error parameters, training data and label pairs respectively.

[0095] If the first layer is a convolutional layer and a pooling layer, the error of the lth layer is:

[0096]

[0097] Where k represents the filter number, Represents the derivative of the activation function itself in the model.

[0098] The gradients of the parameters are:

[0099]

[0100] In the formula, the input of the lth layer is The convolution between the input filter and the error is Similarly, the optimized results are substituted into the neural network models of the sparrow algorithm (SSA) and the BiLSTM neural network prediction model to improve the model training speed and reduce the network parameters of the model.

[0101] The calculation is as follows:

[0102]

[0103] Where: is the forward output of the hidden layer in the up-climbing stage and the down-climbing stage, is the backward output of the hidden layer in the up-climbing stage and the down-climbing stage; L1 indicates forward calculation, and L2 indicates backward calculation. t is the predicted value of the up-climbing stage and the down-climbing stage obtained at time t, c t is the bias optimization parameter at time t; a t represents the forward output weight; b t Represents the backward output weight; i t represents the input of the neural network model of the sparrow algorithm (SSA) and the BiLSTM neural network prediction model at time t; B t ' represents the wind power prediction result; n is the number of calculations.

[0104] After the step 4: using the error correction model to correct the three prediction results to obtain the final wind power prediction result, the method further includes:

[0105] Construct a prediction error sequence based on the wind power prediction value of the above integrated prediction model;

[0106] Predict the prediction error sequence to obtain the error sequence prediction value;

[0107] The error prediction value is combined with the wind power prediction value to obtain the final wind power prediction result after the difference is correct.

[0108] The present invention proposes an error prediction model to learn the error of the integrated prediction model. The error prediction model proposed in the present invention can directly explore the potential relationship between the predicted power and the actual power of the wind power integrated prediction model without the help of error distribution. First, a prediction error sequence is constructed based on the wind power prediction value of the above integrated prediction model; secondly, the error sequence is predicted to obtain the error sequence prediction value; finally, the error prediction value is combined with the wind power prediction value to obtain the final prediction result, and the error correction satisfies the following formula:

[0109] e=B t -B t ' (17)

[0110] e'(t)=f[e(tT),e(t-2T),…,e(t-nT)] (18)

[0111] B t =B t '+e' (19)

[0112] Where: B t is the actual value of wind power; B t' is the preliminary forecast value of wind power; e is the calculated forecast error sequence; e(t-nT) is the nth historical forecast error before time t; e'(t) is the constructed error sequence; B t ” is the final wind power prediction result after error correction.

[0113] Embodiment 2:

[0114] On the other hand, the present invention also proposes a wind power prediction system considering ramp feature recognition and error correction, comprising:

[0115] A ramp classification module is used to identify the ramp events of the historical wind power sequence according to the characteristics; based on the identification results, all wind power sequences are classified into an up-climbing stage, a down-climbing stage and a non-climbing stage;

[0116] Meteorological factor acquisition module, used to obtain meteorological factors of wind power;

[0117] A prediction result determination module is used to establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, and input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction to obtain three prediction results;

[0118] The final wind power prediction result determination module is used to correct the three prediction results using the error correction model to obtain the final wind power prediction result.

[0119] Those skilled in the art will appreciate that the embodiments of the present invention application can be provided as methods, systems, or computer program products. Therefore, the present invention application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present invention application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention application rather than to limit its protection scope. Although the present invention application is described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present invention application, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A wind power prediction method considering ramp feature recognition and error correction, characterized in that: include: Step 1: Identify the ramp events of the historical wind power sequence according to their characteristics; Based on the identification result, all wind power sequences are classified into an up-climbing phase, a down-climbing phase and a non-climbing phase; Step 2: Obtain meteorological factors of wind power; Step 3: Establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction, and obtain three prediction results; Step 4: Use the error correction model to correct the three prediction results to obtain the final wind power prediction result.

2. The method according to claim 1, characterized in that In step 3, prediction models are established for the up-climbing stage, the down-climbing stage, and the non-climbing stage, respectively, including: The SSA-BiLSTM neural network prediction model is used to predict the climbing stage in the wind power sequence, the CNN-BiLSTM neural network prediction model is used to predict the downclimbing stage in the wind power sequence, and the LSTM neural network prediction model is used to predict the non-climbing stage in the wind power sequence.

3. The method according to claim 2, characterized in that In step 1, the features include four, namely, amplitude change ΔP r , duration Δt, climbing rate ΔP r / Δt, climbing direction.

4. The method according to claim 1, characterized in that: In the step 1, based on the identification result, all wind power sequences are classified into an up-climbing phase, a down-climbing phase and a non-climbing phase, including: The plurality of wind power sequences are compressed according to the parallelogram rule; the up-climbing stage, the down-climbing stage and the non-climbing stage are used as constraints of the revolving door algorithm, and parallelograms are continuously constructed so that each of the wind power sequences is contained in the parallelogram; if there are points corresponding to the wind power sequences that are not contained in the parallelogram, then all points before the points corresponding to the currently calculated wind power sequence need to be compressed, and then the long wind power sequence corresponding to the compressed points is divided into an up-climbing sequence, a down-climbing sequence and a non-climbing sequence.

5. The method according to claim 1, characterized in that After obtaining the meteorological factors of wind power, the method further includes: Based on the correlation coefficient, a Pearson correlation coefficient algorithm or a Spearman correlation coefficient algorithm is used to calculate the correlation coefficient between each meteorological factor in each time period and the uphill climbing stage, the downhill climbing stage or the non-climbing stage; Based on the correlation coefficient, meteorological factors having a strong correlation with the uphill climbing stage, the downhill climbing stage or the non-climbing stage are screened out.

6. The method according to claim 5, characterized in that The meteorological factors having a strong correlation with the uphill climbing stage, the downhill climbing stage or the non-climbing stage are screened out based on the correlation coefficient, including screening meteorological factors according to the principle that the closer the correlation coefficient is to 1, the stronger the correlation is.

7. The method according to claim 1, characterized in that The step 4: using the error correction model to correct the three prediction results to obtain the final wind power prediction result, including: For the prediction results corresponding to the up-climbing stage and the down-climbing stage, the power prediction is performed using the neural network model of the sparrow algorithm and the BiLSTM neural network prediction model; The calculation is as follows: Where: is the forward output of the hidden layer in the up-climbing stage and the down-climbing stage, is the backward output of the hidden layer in the up-climbing stage and the down-climbing stage; L1 indicates forward calculation, and L2 indicates backward calculation. t is the predicted value of the up-climbing stage and the down-climbing stage obtained at time t, c t is the bias optimization parameter at time t; a t represents the forward output weight; b t Represents the backward output weight; i t represents the input of the model at time t; B t ' represents the wind power prediction result; n is the number of calculations.

8. The method according to claim 1, characterized in that After the step 4: using the error correction model to correct the three prediction results to obtain the final wind power prediction result, the method further includes: Construct a prediction error sequence based on the wind power prediction value of the above integrated prediction model; Predict the prediction error sequence to obtain the error sequence prediction value; The error prediction value is combined with the wind power prediction value to obtain the final wind power prediction result after the difference is correct.

9. The method according to claim 8, characterized in that The error correction satisfies the following formula: e=B t -B t ' e'(t)=f[e(tT),e(t-2T),…,e(t-nT)] B t ”=B t '+e' Where: B t is the actual value of wind power; B t ' is the preliminary forecast value of wind power; e is the calculated forecast error sequence; e(t-nT) is the nth historical forecast error before time t; e'(t) is the constructed error sequence; B t ” is the final wind power prediction result after error correction.

10. A wind power prediction system considering ramp feature recognition and error correction, comprising: The ramp classification module is used to identify the ramp events of the historical wind power sequence according to their characteristics; Based on the identification result, all wind power sequences are classified into an up-climbing phase, a down-climbing phase and a non-climbing phase; Meteorological factor acquisition module, used to obtain meteorological factors of wind power; A prediction result determination module is used to establish prediction models for the uphill climbing stage, downhill climbing stage and non-climbing stage respectively, and input the identified uphill climbing stage, downhill climbing stage and non-climbing stage and meteorological factors into the corresponding prediction models for prediction to obtain three prediction results; The final wind power prediction result determination module is used to correct the three prediction results using the error correction model to obtain the final wind power prediction result.

Images