Bi-LSTM-QR-based new energy flexibility demand quantification method and system

Abstract

The invention relates to the technical field of flexibility research of a novel power system, in particular to a Bi-LSTM-QR-based new energy flexibility demand quantification method and a Bi-LSTM-QR-based new energy flexibility demand quantification system. The method comprises the steps of long-sequence new energy prediction output and corresponding real output data collection, confidence degree setting and upper and lower bound quantile loss function determination, Bi-LSTM model training, prediction output and fluctuation interval quantization accuracy evaluation, and flexibility demand calculation. Based on the idea of data driving, the flexibility demand of new energy is quantified by using a neural network tool from the perspective of interval probability prediction. The method comprises the following steps: collecting massive long series data based on an existing prediction system of a power grid, embedding a quantile loss function into a neural network Bi-LSTM for training, and fitting a predicted value and an actual quantile mapping relation; and inputting a predicted value into the trained model to obtain a quantile of actual output, converting model output into an actual output fluctuation interval under confidence coefficient based on a central interval prediction method, and finally calculating upper and lower flexibility requirements at each moment based on the interval.

Description

Technical Field

[0001] The present invention relates to the technical field of novel power system flexibility research, and in particular to a new energy flexibility demand quantification method and system based on Bi-LSTM-QR. Background Art

[0002] In recent years, a new power system with a high proportion of new energy, mainly wind power and photovoltaic power, has gradually taken shape. After large-scale wind power and photovoltaic power are connected to the grid, their natural random volatility has caused a considerable impact on the entire power system. Relying on flexible resources such as hydropower and energy storage to smooth out wind and solar fluctuations is an effective means of grid regulation. However, with the rapid growth of new energy installed capacity, its surging flexibility demand at certain times will lead to a serious lack of flexibility on the power grid supply side. Accurately quantifying the flexibility demand of new energy is conducive to assisting the power grid in reasonably arranging backup, which can effectively reduce the probability of insufficient flexibility in the power system.

[0003] At present, the grid side considers the flexibility demand for wind power and photovoltaic power from the perspective of forecast error. For example, the maximum error coefficient is directly given or the probability distribution curve of the error is fitted based on historical data, and the fluctuation boundary of the forecast value of new energy is depicted on this basis to calculate the flexibility demand. Such methods are too simple and fail to make full use of the massive historical data.

[0004] Since the power grid side is equipped with a new energy power prediction system, and with the operation of the system, a large amount of real operation and prediction data has been accumulated. At the same time, the development of machine learning, neural networks, etc. provides an effective means for massive data processing. To this end, the present invention proposes a Bi-LSTM-QR new energy flexibility demand quantification method based on data-driven thinking. The quantile loss function is embedded in the neural network Bi-LSTM for training, and the quantile mapping relationship between the predicted value and the actual output is fitted. After the predicted value is input into the trained model, the quantile of the actual output can be obtained. Based on the central interval prediction method, the model output is converted into an actual output fluctuation interval under a certain confidence level. Finally, the first-order difference of adjacent time periods is calculated to obtain the new energy flexibility demand of each time period. The actual data of a power grid in the southwest is used for verification, which shows the effectiveness of the quantification method. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is that the sharp increase in flexibility demand at certain moments will lead to a serious lack of flexibility on the power grid supply side. Accurately quantifying the flexibility demand of new energy sources is conducive to assisting the power grid in rationally arranging backup, and can effectively reduce the probability of insufficient flexibility in the power system. At present, the flexibility demand of wind power and photovoltaic power on the power grid side is considered from the perspective of prediction error. For example, the maximum error coefficient is directly given or a probability distribution curve based on the error of fitting historical data is used to characterize the fluctuation boundary of the new energy prediction value based on this, so as to calculate the flexibility demand. This type of method is too simple and fails to make full use of the massive amount of historical data.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a method for quantifying the flexibility demand of new energy based on Bi-LSTM-QR, which includes collecting long-sequence predicted output and actual output data of new energy to form a training set and a test set;

[0009] Set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training;

[0010] Use the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range;

[0011] The reliability of prediction intervals was evaluated using interval coverage and normalized mean interval width;

[0012] Based on the predicted actual output fluctuation range, the upward and downward flexibility requirements of renewable energy are calculated.

[0013] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method described in the present invention, the long-sequence new energy predicted output and actual output data are collected to form a training set and a test set, including the collection of long-sequence new energy predicted output and the corresponding actual output data, relying on the power grid prediction system, extracting the long-sequence new energy predicted output and the corresponding actual output to form an original data set, converting the original data set into unit installed capacity, and then dividing the data set into a training set and a test set according to a ratio of 8:2.

[0014] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method of the present invention, wherein: the confidence is set, the confidence setting and the upper and lower quantile loss functions are determined, the confidence is set to (1-β), and the upper and lower quantile determination formulas are as follows:

[0015]

[0016] in, τis the lower quantile, is the upper quantile. When the confidence level is set to 90%, that is, β is taken as 0.1, the corresponding upper quantile is the 95% quantile and the lower quantile is the 5% quantile.

[0017] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method of the present invention, wherein: the quantile regression loss function is constructed, the collected data samples are composed of historical predicted output and actual output, and the corresponding quantile regression loss function can be adopted as follows:

[0018]

[0019] Among them, there are D days of actual dispatching operation data, and the number of output data collection times per day is K, so D×K groups of samples can be obtained in It is the D day of history. is the predicted output and actual output of the kth observation point. For a certain quantile τ, ρ τ is the pinball loss function, which uses the following formula:

[0020]

[0021] Among them, τ represents a certain quantile, and μ represents the difference between the actual value and the predicted value.

[0022] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method of the present invention, wherein: based on the set confidence level, a quantile regression loss function is constructed and embedded in the Bi-LSTM model for training;

[0023] The constructed quantile regression loss function is embedded in the Bi-LSTM model training. The training is to continuously optimize and adjust the weight matrix and bias matrix W(τ), V(τ) to minimize the quantile regression loss function, so as to enable the neural network to learn the nonlinear implicit relationship between the massive input prediction value and the actual output under different quantile conditions. The long short-term memory network is used to solve the long-term dependence problem of the recurrent neural network in the training process. The Bi-LSTM based on bidirectional information can fully mine the information before and after the sequence. Its output is jointly determined by forward propagation and back propagation, and the fluctuation range is obtained at the same time, which uses the following expression:

[0024] H t =[h f,t ,h b,t ]

[0025] Among them, h f，t is the forward long short-term memory network propagation layer, h b，tis the output of the reverse long short-term memory network propagation layer.

[0026] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method of the present invention, wherein: the reliability of the prediction interval is evaluated based on the interval coverage and the normalized average interval width, and the accuracy evaluation of the prediction output and the fluctuation interval quantification adopts the following steps:

[0027] Use the trained model to predict the upper and lower quantiles corresponding to the test set, construct the fluctuation range of the actual output, and use the following indicators to evaluate the reliability of the range;

[0028] Reliability index - interval coverage, which represents the probability that the future true value falls within the interval. When the prediction interval includes all extreme cases of the target, the coverage probability will be 1. The calculation can be done using the following formula:

[0029]

[0030] Among them, N is the total number of data in the test set, ξ is the total number of real outputs falling within the interval;

[0031] Sensitivity index - normalized average interval width. The smaller the average interval width value, the stronger the model's ability to describe power uncertainty. The calculation can be done using the following formula:

[0032]

[0033] Among them, σ i is the normalized interval width of time period i.

[0034] As a preferred solution of the Bi-LSTM-QR-based new energy flexibility demand quantification method of the present invention, wherein: the upward and downward flexibility demands of new energy are calculated based on the predicted real output fluctuation range, wherein the coupling relationship of time is considered, and the upward flexibility demand and downward flexibility demand of new energy can be calculated using the following formula:

[0035]

[0036] in, To meet the upward flexibility demand of new energy at time t, is the downward flexibility demand of the new energy at time t, P t f is the predicted value of new energy at time t, is the upper bound of the actual output fluctuation at time t+1, It is the lower bound of the actual output fluctuation at time t+1, which is obtained from the interval prediction result in the previous step.

[0037] In a second aspect, an embodiment of the present invention provides a new energy flexibility demand quantification system based on Bi-LSTM-QR, which includes

[0038] The data collection module collects long-sequence predicted and actual output data of new energy sources to form training and test sets;

[0039] Build the module, set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training;

[0040] The prediction module uses the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range;

[0041] Evaluation module, which uses interval coverage and normalized average interval width to evaluate the reliability of prediction intervals;

[0042] The output module calculates the upward and downward flexibility requirements of renewable energy based on the predicted actual output fluctuation range.

[0043] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the new energy flexibility demand quantification method based on Bi-LSTM-QR as described in the first aspect of the present invention are implemented.

[0044] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the new energy flexibility demand quantification method based on Bi-LSTM-QR as described in the first aspect of the present invention are implemented.

[0045] The beneficial effects of the present invention are: providing a more sophisticated method for quantifying the flexibility demand of new energy, utilizing massive historical data, based on data-driven thinking, using neural networks to fit the nonlinear implicit relationship between different quantiles of the predicted output of new energy and the actual output, introducing the idea of ​​interval probability prediction, using the predicted value as the feature input, and obtaining a reliable fluctuation range of the actual output, providing corresponding guidance for the reasonable configuration of power grid scheduling, especially the standby. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1Flow chart of the new energy flexibility demand quantification method based on Bi-LSTM-QR;

[0048] Figure 2 Computer equipment diagram for the new energy flexibility demand quantification method based on Bi-LSTM-QR;

[0049] Figure 3 This is the Bi-LSTM model structure diagram of the new energy flexibility demand quantification method based on Bi-LSTM-QR;

[0050] Figure 4 This is the Bi-LSTM model structure diagram of the new energy flexibility demand quantification method based on Bi-LSTM-QR. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0054] Example 1

[0055] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a new energy flexibility demand quantification method based on Bi-LSTM-QR, including:

[0056] S100: Collect long-sequence predicted and actual output data of new energy sources to form training and test sets;

[0057] S101: The long-sequence predicted output and actual output data of new energy are collected to form a training set and a test set, including the collection of long-sequence predicted output of new energy and the corresponding actual output data. Relying on the power grid prediction system, the long-sequence predicted output of new energy and the corresponding actual output are extracted to form an original data set. After the original data set is converted into unit installed capacity, the data set is divided into a training set and a test set according to a ratio of 8:2.

[0058] S200: Set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training;

[0059] S201: The confidence is set, and the confidence setting and the upper and lower quantile loss functions are determined. The confidence is set to (1-β), and the upper and lower quantile determination formulas are as follows:

[0060]

[0061] Among them, τ is the lower bound quantile, is the upper quantile. When the confidence level is set to 90%, that is, β is taken as 0.1, the corresponding upper quantile is the 95% quantile and the lower quantile is the 5% quantile.

[0062] S202: The quantile regression loss function is constructed. The collected data samples are composed of historical predicted output and actual output. The corresponding quantile regression loss function can be expressed as follows:

[0063]

[0064] Among them, there are D days of actual dispatching operation data, and the number of output data collection times per day is K, so D×K groups of samples can be obtained in It is the D day of history. is the predicted output and actual output of the kth observation point. For a certain quantile τ, ρ τ is the pinball loss function, which uses the following formula:

[0065]

[0066] Among them, τ represents a certain quantile, and μ represents the difference between the actual value and the predicted value.

[0067] S300: Use the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range;

[0068] S301: Based on the set confidence level, a quantile regression loss function is constructed and embedded into a Bi-LSTM model for training;

[0069] The constructed quantile regression loss function is embedded in the Bi-LSTM model training. The training is to continuously optimize and adjust the weight matrix and bias matrix W(τ), V(τ) to minimize the quantile regression loss function, so as to enable the neural network to learn the nonlinear implicit relationship between the massive input prediction value and the actual output under different quantile conditions. The long short-term memory network is used to solve the long-term dependence problem of the recurrent neural network in the training process. The Bi-LSTM based on bidirectional information can fully mine the information before and after the sequence. Its output is jointly determined by forward propagation and back propagation, and the fluctuation range is obtained at the same time, which uses the following expression:

[0070] H t =[h f,t ,h b,t ]

[0071] Among them, h f，t is the forward long short-term memory network propagation layer, h b，t is the output of the reverse long short-term memory network propagation layer.

[0072] S400: Use interval coverage and normalized mean interval width to assess the reliability of prediction intervals;

[0073] S401: The reliability of the prediction interval is evaluated based on the interval coverage and the normalized average interval width, and the accuracy evaluation of the prediction output and the fluctuation interval quantification is performed using the following steps:

[0074] Use the trained model to predict the upper and lower quantiles corresponding to the test set, construct the fluctuation range of the actual output, and use the following indicators to evaluate the reliability of the range;

[0075] Reliability index - interval coverage, which represents the probability that the future true value falls within the interval. When the prediction interval includes all extreme cases of the target, the coverage probability will be 1. The calculation can be done using the following formula:

[0076]

[0077] Among them, N is the total number of data in the test set, ξ is the total number of real outputs falling within the interval;

[0078] Sensitivity index - normalized average interval width. The smaller the average interval width value, the stronger the model's ability to describe power uncertainty. The calculation can be done using the following formula:

[0079]

[0080] Among them, σ i is the normalized interval width of time period i.

[0081] S500: Calculates the upward and downward flexibility requirements of new energy based on the predicted actual output fluctuation range.

[0082] S501: Based on the predicted real output fluctuation range, the upward and downward flexibility requirements of the new energy are calculated, wherein the coupling relationship of time is considered, and the upward flexibility requirements and downward flexibility requirements of the new energy can be calculated using the following formula:

[0083]

[0084] in, To meet the upward flexibility demand of new energy at time t, is the downward flexibility demand of the new energy at time t, P t f is the predicted value of new energy at time t, is the upper bound of the actual output fluctuation at time t+1, It is the lower bound of the actual output fluctuation at time t+1, which is obtained from the interval prediction result in the previous step.

[0085] Furthermore, this embodiment also provides a new energy flexibility demand quantification system based on Bi-LSTM-QR, including:

[0086] The data collection module collects long-sequence predicted and actual output data of new energy sources to form training and test sets;

[0087] Build the module, set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training;

[0088] The prediction module uses the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range;

[0089] Evaluation module, which uses interval coverage and normalized average interval width to evaluate the reliability of prediction intervals;

[0090] The output module calculates the upward and downward flexibility requirements of renewable energy based on the predicted actual output fluctuation range.

[0091] This embodiment also provides a computer device, which is applicable to the case of a new energy flexibility demand quantification method based on Bi-LSTM-QR, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the new energy flexibility demand quantification method based on Bi-LSTM-QR as proposed in the above embodiment.

[0092] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0093] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for quantifying new energy flexibility demand based on Bi-LSTM-QR as proposed in the above embodiment is implemented.

[0094] In summary, a more sophisticated method for quantifying the flexibility demand of new energy is provided. Based on the data-driven idea, a neural network is used to fit the nonlinear implicit relationship between the predicted output of new energy and different quantiles of the actual output based on the historical massive data. The idea of ​​interval probability prediction is introduced, and the predicted value is used as the feature input to obtain the reliable fluctuation range of the actual output, which provides corresponding guidance for the reasonable configuration of power grid dispatch, especially the reserve.

[0095] This method first collects a long sequence of new energy predicted output and actual output data to construct a training set and a test set. Then, the confidence level is set, and the quantile regression loss function is constructed. It is embedded in the bidirectional long short-term memory (Bi-LSTM) neural network model for training. The nonlinear relationship between the predicted value and the true value at different quantiles is fitted by minimizing the loss function. The trained model is used to predict the upper and lower quantiles of the actual output, thereby constructing a fluctuation range under a certain confidence level. The interval coverage and normalized mean width are further used to evaluate the reliability and sensitivity of the prediction interval. Finally, based on the predicted interval, the flexibility requirements for upward and downward adjustment of new energy are calculated.

[0096] Compared with traditional methods, this invention is based on massive historical data, introduces the idea of ​​interval probability prediction, and uses advanced deep learning models to characterize the implicit connection between predicted values ​​and true values, which can obtain a more accurate and reliable true output fluctuation range under a certain confidence level. On this basis, the flexibility demand of new energy is quantified to provide important reference information for power grid dispatching and standby optimization configuration, which can avoid waste of resources caused by blind conservatism in dispatching decisions. Simulation experiments show that compared with other methods, the proposed Bi-LSTM-QR model can obtain the narrowest prediction interval while ensuring a high confidence level, which verifies that this method can estimate the fluctuation range of new energy output with minimal uncertainty, which is of great significance to improving the economy of power grid dispatching.

[0097] The long-sequence prediction output of new energy and the corresponding real output data collection, confidence setting and upper and lower quantile loss function determination, Bi-LSTM model training, prediction output and fluctuation range quantification accuracy evaluation, and flexibility demand calculation. The present invention is based on the idea of ​​data-driven, starting from the perspective of interval probability prediction, and uses neural network tools to quantify the flexibility demand of new energy. First, based on the existing prediction system of the power grid, a large amount of long-series data is collected, and the quantile loss function is embedded in the neural network Bi-LSTM for training, and the quantile mapping relationship between the predicted value and the actual output is fitted. After the predicted value is input into the trained model, the quantile of the actual output can be obtained. Based on the central interval prediction method, the model output is converted into the actual output fluctuation range under a certain confidence level. Finally, based on this interval, the upper and lower flexibility requirements at each moment are calculated, which can provide relevant guidance for the reasonable backup of the power grid.

[0098] Example 2

[0099] Reference Figure 1 - Figure 4 , which is the second embodiment of the present invention, and this embodiment provides a new energy flexibility demand quantification method based on Bi-LSTM-QR. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0100] Based on the 15-minute predicted output and the corresponding actual output data collected by the wind power prediction system of a power grid in southwest my country for one year, and combined with the traditional error distribution-based method on the power grid side, a comparison of interval quantification accuracy is performed to verify the effectiveness of the present invention.

[0101] Figure 1 The overall flow chart of this method is as follows: Figure 3It is a structural diagram of the Bi-LSTM model. Take setting the confidence interval of 80% and 90% as an example. In the present invention, the upper and lower bounds of the 80% confidence interval are the 90% quantile and the 10% quantile, respectively, and the upper and lower bounds of the 90% confidence interval are the 95% quantile and the 5% quantile, respectively. The number of neurons of the Bi-LSTM model is set to 64, the time step is set to 24, the learning rate is set to 0.001, and the number of iterations is set to 100.

[0102] Figure 4 The frequency distribution histogram of the prediction error is obtained by fitting the distribution with a Gaussian mixture distribution containing two sub-distributions (denoted as GMM2) and a Gaussian mixture distribution containing three sub-distributions (denoted as GMM3). The interval between 0.1 and 0.9 of the cumulative frequency curve is used as the 80% confidence interval, and the interval between 0.05 and 0.95 of the cumulative frequency curve is used as the 90% confidence interval. At the same time, the traditional quantile linear regression (denoted as QR) is added for comparison. The interval quantification results of each model are shown in the following table.

[0103] Table 1 Quantitative results of each model

[0104]

[0105] From the perspective of the PICP index, all methods meet the set confidence requirements, and the PICP values ​​of the quantile linear regression are the highest. However, the PINAW index is not much different from the GMM2 and GMM3 methods. The present invention has the smallest interval width while meeting the confidence level, which also shows that the method of the present invention has the smallest uncertainty. The new energy flexibility demand calculated based on this interval can provide corresponding support for the uncertainty of power grid scheduling and can effectively reduce the standby cost.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A new energy flexibility demand quantification method based on Bi-LSTM-QR, characterized by: include, Collect long-sequence predicted and actual output data of new energy sources to form training and test sets; Set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training; Use the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range; The reliability of prediction intervals was evaluated using interval coverage and normalized mean interval width; Based on the predicted actual output fluctuation range, the upward and downward flexibility requirements of renewable energy are calculated.

2. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 1, characterized in that: The method collects long-sequence new energy predicted output and actual output data to form a training set and a test set, including the collection of long-sequence new energy predicted output and the corresponding actual output data, relying on the power grid prediction system, extracting the long-sequence new energy predicted output and the corresponding actual output to form an original data set, converting the original data set into unit installed capacity, and then dividing the data set into a training set and a test set according to an 8:2 ratio.

3. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 2, characterized in that: The confidence level is set, and the confidence level setting and the upper and lower quantile loss functions are determined. The confidence level is set to (1-β), and the upper and lower quantile determination formulas are as follows: in, is the lower quantile, is the upper quantile. When the confidence level is set to 90%, that is, β is taken as 0.1, the corresponding upper quantile is the 95% quantile and the lower quantile is the 5% quantile.

4. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 3, characterized in that: The quantile regression loss function is constructed, and the collected data samples are composed of historical predicted output and actual output. The corresponding quantile regression loss function can be adopted as follows: Among them, there are D days of actual dispatching operation data, and the number of output data collection times per day is K, so D × K group samples d=1,2,…,D;k=1,2,…,K,where It is the D day of history. is the predicted output and actual output of the kth observation point. For a certain quantile τ, ρ τ is the pinball loss function, which uses the following formula: Among them, τ represents a certain quantile, and μ represents the difference between the actual value and the predicted value.

5. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 4, characterized in that: Based on the set confidence level, a quantile regression loss function is constructed and embedded into the Bi-LSTM model for training; The constructed quantile regression loss function is embedded in the Bi-LSTM model training. The training is to continuously optimize and adjust the weight matrix and bias matrix W(τ), V(τ) to minimize the quantile regression loss function, so as to enable the neural network to learn the nonlinear implicit relationship between the massive input prediction value and the actual output under different quantile conditions. The long short-term memory network is used to solve the long-term dependence problem of the recurrent neural network in the training process. The Bi-LSTM based on bidirectional information can fully mine the information before and after the sequence. Its output is jointly determined by forward propagation and back propagation, and the fluctuation range is obtained at the same time, which uses the following expression: H t =[h f,t ,h b,t ] Among them, h f，t is the forward long short-term memory network propagation layer, h b，t is the output of the reverse long short-term memory network propagation layer.

6. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 5, characterized in that: The reliability of the forecast interval is evaluated based on the use of interval coverage and normalized average interval width, and the accuracy evaluation of the forecast output and the fluctuation interval quantification adopts the following steps: Use the trained model to predict the upper and lower quantiles corresponding to the test set, construct the fluctuation range of the actual output, and use the following indicators to evaluate the reliability of the range; Reliability index - interval coverage, which represents the probability that the future true value falls within the interval. When the prediction interval includes all extreme cases of the target, the coverage probability will be 1. The calculation can be done using the following formula: Among them, N is the total number of data in the test set, ξ is the total number of real outputs falling within the interval; Sensitivity index - normalized average interval width. The smaller the average interval width value, the stronger the model's ability to describe power uncertainty. The calculation can be done using the following formula: Among them, σ i is the normalized interval width of time period i.

7. The method for quantifying new energy flexibility demand based on Bi-LSTM-QR as claimed in claim 6, characterized in that: The upward and downward flexibility requirements of new energy are calculated based on the predicted real output fluctuation range, wherein the coupling relationship of time is considered. The upward flexibility requirements and downward flexibility requirements of new energy can be calculated using the following formula: in, To meet the upward flexibility requirements of new energy at time t, is the downward flexibility demand of the new energy at time t, is the predicted value of new energy at time t, is the upper bound of the actual output fluctuation at time t+1, It is the lower bound of the actual output fluctuation at time t+1, which is obtained from the interval prediction result in the previous step.

8. A new energy flexibility demand quantification system based on Bi-LSTM-QR, based on the new energy flexibility demand quantification method based on Bi-LSTM-QR according to any one of claims 1 to 7, characterized in that: Also includes, The data collection module collects long-sequence predicted and actual output data of new energy sources to form training and test sets; Build the module, set the confidence level, construct the quantile regression loss function, and embed it into the Bi-LSTM model for training; The prediction module uses the trained Bi-LSTM model to predict the upper and lower bounds of the actual output and construct the fluctuation range; Evaluation module, which uses interval coverage and normalized average interval width to evaluate the reliability of prediction intervals; The output module calculates the upward and downward flexibility requirements of renewable energy based on the predicted actual output fluctuation range.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the new energy flexibility demand quantification method based on Bi-LSTM-QR as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the new energy flexibility demand quantification method based on Bi-LSTM-QR as described in any one of claims 1 to 7 are implemented.

Images