Large-scale new energy short-term power prediction method and device in extreme weather

Abstract

The invention provides a large-scale new energy short-term power prediction method and device in extreme weather, and relates to the technical field of new energy, and the method comprises the steps: judging whether a current time period is an extreme weather scene or not according to an extreme weather detection network, and calculating a scene abrupt change rate if the current time period is the extreme weather scene; establishing an influence relation curve of the scene mutation rate sample and the power prediction sample, and analyzing the scene mutation rate; inputting the power prediction influence index into a short-term power prediction window regulator, and outputting a phase spread ratio corresponding to the power prediction influence index; and performing phase expansion on the time period in the extreme weather scene, calling the short-term power prediction model to perform power prediction according to the expanded time phase, and outputting a power prediction result. According to the invention, the technical effects of improving the response sensitivity and fitting precision of short-term power prediction in extreme weather and guaranteeing the stable operation and scheduling efficiency of a power grid can be achieved.

Description

Technical Field

[0001] This application relates to the field of new energy technologies, and particularly to a method and device for short-term power prediction of large-scale new energy under extreme weather conditions. Background Art

[0002] In the context of the rapid development of new energy power generation, the proportion of renewable energy sources such as wind power and photovoltaic power in the power grid is increasing day by day, and the operational stability and dispatching flexibility of the power system are facing unprecedented challenges. Due to the characteristics of strong volatility, large randomness, and poor predictability of such new energy sources, traditional power prediction methods are no longer able to meet the actual operational requirements.

[0003] Existing short-term power prediction technologies mainly rely on time series modeling of historical data or learning of static meteorological factors, and their prediction periods are mostly set to a fixed length, such as making a prediction every 10 minutes or 15 minutes. This prediction mechanism with a fixed time window can provide relatively reliable prediction results in the case of stable or slowly changing weather. However, once extreme weather scenarios are encountered, such as sudden weather conditions like strong wind, heavy rain, thunderstorms, high temperature, or low temperature, the response ability and fitting accuracy of the prediction model will significantly decline, resulting in a substantial increase in prediction errors, ultimately affecting the safety and economy of power grid dispatching.

[0004] In summary, there is a technical problem in the existing technology that due to the fixed prediction period and the inability to dynamically adjust according to real-time weather changes, the model response lags and the fitting accuracy decreases in extreme weather scenarios, further affecting the accuracy of new energy power prediction and the safety and economy of power grid dispatching. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for short-term power prediction of large-scale new energy under extreme weather conditions, so as to solve the technical problem in the existing technology that due to the fixed prediction period and the inability to dynamically adjust according to real-time weather changes, the model response lags and the fitting accuracy decreases in extreme weather scenarios, further affecting the accuracy of new energy power prediction and the safety and economy of power grid dispatching.

[0006] In view of the above problems, this application provides a method and device for short-term power prediction of large-scale new energy under extreme weather conditions.

[0007] In a first aspect, the present application provides a method for short-term power prediction of large-scale new energy under extreme weather conditions, which is implemented through a short-term power prediction device for large-scale new energy under extreme weather conditions, and includes: constructing an extreme weather detection network, determining whether the current time period is an extreme weather scenario according to the extreme weather detection network, and calculating a scenario mutation rate if the current time period is an extreme weather scenario, where the scenario mutation rate is the mutation rate of the extreme weather scenario in the current time period compared to the corresponding scenario in the previous time period; establishing an influence relationship curve between the scenario mutation rate sample and the power prediction sample, analyzing the scenario mutation rate according to the influence relationship curve to obtain a power prediction influence index; inputting the power prediction influence index into a short-term power prediction window regulator to output a phase expansion ratio corresponding to the power prediction influence index; expanding the phase of the time period under the extreme weather scenario according to the phase expansion ratio to obtain an expanded extended time phase, and calling a short-term power prediction model to perform power prediction according to the extended time phase, and outputting a power prediction result.

[0008] In a second aspect, the present application further provides a short-term power prediction device for large-scale new energy under extreme weather conditions, which is used to execute the short-term power prediction method for large-scale new energy under extreme weather conditions as described in the first aspect, and includes: a judgment module, configured to construct an extreme weather detection network, determine whether the current time period is an extreme weather scenario according to the extreme weather detection network, and calculate a scenario mutation rate if the current time period is an extreme weather scenario, where the scenario mutation rate is the mutation rate of the extreme weather scenario in the current time period compared to the corresponding scenario in the previous time period; an analysis module, configured to establish an influence relationship curve between the scenario mutation rate sample and the power prediction sample, analyze the scenario mutation rate according to the influence relationship curve to obtain a power prediction influence index; an output module, configured to input the power prediction influence index into a short-term power prediction window regulator to output a phase expansion ratio corresponding to the power prediction influence index; a prediction module, configured to expand the phase of the time period under the extreme weather scenario according to the phase expansion ratio to obtain an expanded extended time phase, and call a short-term power prediction model to perform power prediction according to the extended time phase, and output a power prediction result.

[0009] The technical solution provided in the present application has at least the following technical effects or advantages: By achieving the technical goal of dynamically adjusting the power prediction time phase window based on extreme weather detection and scenario mutation rate analysis, the technical effects of improving the response sensitivity and fitting accuracy of short-term power prediction under extreme weather conditions and ensuring the stable operation and dispatching efficiency of the power grid are achieved.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specific embodiments of this application are specifically given. It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of this application, nor is it used to limit the scope of this application. Other features of this application will become easily understandable through the following description of the specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0012] Figure 1 It is a schematic flowchart of the method for short-term power prediction of large-scale new energy under extreme weather conditions of this application; Figure 2 It is a schematic structural diagram of the device for short-term power prediction of large-scale new energy under extreme weather conditions of this application.

[0013] Description of reference numerals: judgment module 11, analysis module 12, output module 13, prediction module 14. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] By providing a method and device for short-term power prediction of large-scale new energy under extreme weather conditions, this application solves the technical problem in the prior art that due to the fixed prediction period and the inability to dynamically adjust according to real-time weather changes, the model response lags and the fitting accuracy decreases in extreme weather scenarios, further affecting the accuracy of new energy power prediction and the safety and economy of power grid dispatching. The technical goal of dynamically adjusting the power prediction time phase window based on extreme weather detection and scenario mutation rate analysis is achieved, and the technical effect of improving the response sensitivity and fitting accuracy of short-term power prediction under extreme weather and ensuring the stable operation and dispatching efficiency of the power grid is achieved.

[0015] Next, the technical solutions in this application will be clearly and completely described with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited by the exemplary embodiments described here. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application. Additionally, it should be noted that for the sake of convenience of description, only the parts related to this application are shown in the drawings rather than all of them.

[0016] Example 1. Please refer to the appendix Figure 1 , this application provides a short-term power prediction method for large-scale new energy under extreme weather, which is applied to a short-term power prediction device for large-scale new energy under extreme weather, and specifically includes the following steps: S1: Construct an extreme weather detection network, and determine whether the current period is an extreme weather scenario according to the extreme weather detection network. If the current period is an extreme weather scenario, calculate the scenario mutation rate, where the scenario mutation rate is the mutation rate of the extreme weather scenario in the current period compared to the corresponding scenario in the previous period.

[0017] Specifically, constructing an extreme weather detection network means establishing an artificial intelligence model to determine whether extreme weather has occurred. Extreme weather includes heavy rain, heavy snow, typhoons, strong winds, high temperatures, cold snaps, etc., which have a significant impact on new energy power generation. The detection network is an algorithm system that can identify these weather states and can automatically learn and identify patterns of abnormal meteorological conditions by using a large amount of historical meteorological data and real-time data. The purpose of constructing this network is to be able to identify in a timely manner when the weather changes drastically, which helps to improve the stability and accuracy of new energy power prediction.

[0018] Determine whether the current period is an extreme weather scenario according to the extreme weather detection network, and receive the current meteorological characteristic data in real time, such as temperature, wind speed, humidity, and atmospheric pressure, and input them into the extreme weather detection network. The extreme weather detection network extracts and analyzes the features of the input data through multiple parameters, and then outputs the scenario mutation rate to determine whether there is an extreme weather phenomenon in this period. The scenario mutation rate represents a numerical value of the weather change amplitude, reflecting the severity of the change in meteorological conditions between adjacent periods.

[0019] The scenario mutation rate refers to the degree of change between the extreme weather scenario in the current period and the corresponding scenario in the previous period, reflecting the change in weather state from the previous period to the current period. If the meteorological characteristics change drastically between two periods, then the scenario mutation rate will be higher, otherwise it will be lower.

[0020] S2: Establish an influence relationship curve between the scenario mutation rate sample and the power prediction sample, and analyze the scenario mutation rate according to the influence relationship curve to obtain a power prediction influence index.

[0021] Specifically, by collecting the degree of weather mutation and the corresponding power prediction results for each time period in historical data, statistical modeling is performed using data samples to form a function curve describing the relationship between the two. The scene mutation rate sample represents the amplitude of the change in meteorological parameters for each time period, such as the change rates of temperature, humidity, wind speed, etc.; the power prediction sample is the estimated value of the new energy power generation by the prediction model during these time periods. Through linear or non-linear fitting methods, a mathematical model can be obtained to describe the influence law of drastic weather changes on power prediction.

[0022] Analyze the scene mutation rate according to the influence relationship curve, input the current scene mutation rate into the previously established influence relationship model, so as to evaluate the trend of prediction error brought by weather fluctuations, obtain the power prediction influence index, which is used to measure the influence degree of the current scene mutation on the power prediction accuracy, and is used to prompt whether it is necessary to adjust the prediction strategy, shorten the time window or introduce redundant compensation. The larger the value of the power prediction influence index, the wider the possible range of prediction error, and more cautious handling of the scheduling and control operations for this time period should be carried out.

[0023] S3: Input the power prediction influence index into the short-term power prediction window regulator, and output the phase expansion ratio corresponding to the power prediction influence index.

[0024] Specifically, input the power prediction influence index into the short-term power prediction window regulator. That is, after detecting the influence degree of the current weather mutation on power prediction, the value of the influence degree is transmitted to a module responsible for adjusting the prediction window. The short-term power prediction window regulator is a dynamic adjustment mechanism that can change the time range referred to by the model according to different prediction risks. By shortening the time window, it can respond more sensitively to meteorological fluctuations and improve the timeliness and accuracy of prediction.

[0025] Then, output the phase expansion ratio corresponding to the power prediction influence index, that is, according to the current risk index, return a corresponding scaling ratio, which is used to determine the time step of the prediction. The phase expansion ratio is a value less than 1 and is used to control the compression degree of the time window. For example, if the prediction was originally made every 10 minutes, and the phase expansion ratio at this time is 1 / 2, that is, 0.5, it means that the regulator recommends increasing the prediction frequency to once every 5 minutes. This adjustment allows the model to sample and calculate more frequently when the weather changes rapidly, and more sensitively capture the power change trend.

[0026] S4: Expand the time period in the extreme weather scene according to the phase expansion ratio to obtain the expanded extended time phase, call the short-term power prediction model to perform power prediction according to the expanded time phase, and output the power prediction result.

[0027] Specifically, in the case where the current meteorological conditions are identified as extreme weather, the prediction time window is shortened according to the phase expansion ratio to obtain the extended time phase after expansion. The extended time phase refers to the time interval actually used for prediction after applying the phase expansion ratio, which is used to replace the conventional prediction time span, enabling it to respond more quickly to external changes and reduce prediction latency.

[0028] Next, the short-term power prediction model is called to perform power prediction according to the extended time phase, which means that a trained short-term power prediction model will be started or run, taking the extended time phase as an input parameter to guide the model to execute the prediction task according to the new time frequency. The short-term power prediction model is generally based on time series algorithms, neural networks, or hybrid models, and is specifically used to accurately estimate the power output within a short time range. Finally, after prediction based on the extended time phase, the power prediction value corresponding to the time point will be generated and used in system decision-making scenarios such as power grid scheduling, energy storage control, or operation optimization.

[0029] Furthermore, this application also includes: collecting the meteorological feature dimensions of the current period; the extreme weather detection network includes a convolutional layer, a fully connected layer, and an activation output layer. Among them, the convolutional layer is used to perform local change feature convolution on the meteorological feature dimensions, the fully connected layer is used to perform feature fusion on the convolved meteorological feature dimensions, and the activation output layer is used to output the result of whether the current period is an extreme weather scenario according to the fused features.

[0030] Specifically, the meteorological feature dimensions of the current period are collected, and multiple meteorological factors at the current time point are obtained in real time as inputs, including wind speed, wind direction, temperature, humidity, atmospheric pressure, precipitation, solar radiation intensity, etc., and these are sent as inputs to the detection network for judgment.

[0031] The extreme weather detection network consists of three parts: a convolutional layer, a fully connected layer, and an activation output layer. The convolutional layer is a neural network structure that can extract data features in local areas. In meteorological data, it can identify abnormal fluctuations in space or time such as sudden changes in wind speed or temperature. The fully connected layer is used to globally integrate and combine the local features after convolution, enabling the extreme weather detection network to understand the current meteorological state globally. The activation output layer is used to output the judgment result, and through a mathematical function (such as sigmoid or softmax), the fused features are converted into a judgment result of whether it is extreme weather. For example, an output value of 1 represents extreme weather, and 0 represents non-extreme weather. Table 1 shows the record of the most recent judgment of the extreme weather scenario based on the meteorological feature dimensions.

[0032] Table 1: Record of the most recent judgment of the extreme weather scenario based on the meteorological feature dimensions.

[0033]

[0034] Furthermore, the present application also includes: if the current period is an extreme weather scene, the expression for calculating the scene mutation rate includes: ;in, is the scene mutation rate of the current period, is the number of meteorological feature dimensions, is the value of the ith meteorological feature in the current period t, is the value of the ith meteorological feature in the previous period t-1, is a very small positive number used to prevent the denominator from being zero.

[0035] Specifically, if the current period is an extreme weather scenario, it means that the extreme weather detection network has determined that there are extreme weather phenomena in the current period. Extreme weather scenarios indicate that meteorological variables have experienced drastic fluctuations or changes beyond the normal range, which may have a significant impact on the operation of the new energy system, such as abnormal power fluctuations caused by strong winds at wind farms. The scenario mutation rate is calculated using the expression for calculating the scenario mutation rate. The scenario mutation rate is an indicator of the severity of changes in the meteorological environment, which helps the extreme weather detection network determine the severity of weather changes, thereby further adjusting the response speed and sensitivity of the new energy power prediction model.

[0036] is the scene mutation rate of the current period, that is, the calculation target of the expression for calculating the scene mutation rate. The larger the value of the scene mutation rate, the more drastic the weather changes in a short period of time, and vice versa.

[0037] The number of meteorological feature dimensions indicates the total number of meteorological factors monitored and involved in the calculation. If five features, including wind speed, temperature, air pressure, humidity and precipitation, are collected, the number of meteorological feature dimensions is 5. The higher the value of the number of meteorological feature dimensions, the stronger the perception ability of the detection system is, and the more detailed the identification of extreme weather is, and vice versa.

[0038] is the value of the i-th meteorological feature in the current period t and The value of the i-th meteorological feature in the previous period t-1 is used to calculate the relative change percentage as the basis for judging the degree of mutation.

[0039] is a very small positive number used to prevent the denominator from being zero. The minimum value is 0.001 to avoid calculation errors when the value of the meteorological feature in the previous period is 0. For example, the precipitation is 0 at t-1, and becomes 2 mm at t. If the minimum value is not added, it will be divided by 0, causing the program to crash.

[0040] Furthermore, this application also includes: wherein, the scene mutation rate sample is obtained by calculating the scene mutation rate of historical period samples, and the power prediction sample is obtained by calculating the power prediction values of the historical period samples; identifying an error sample set between the power prediction value and the actual power value of the historical period samples; using the vector with the scene mutation rate sample and the power prediction sample as inputs, and using the corresponding error sample set as a label identifying the influence relationship for linear fitting to obtain the influence relationship curve.

[0041] Specifically, the scene mutation rate sample refers to calculating the weather change amplitude index for each period by analyzing meteorological data of multiple historical time periods, which is used to describe the degree of change in meteorological characteristics in different historical scenarios. By analyzing the meteorological characteristics of each past period, such as changes in wind speed, temperature, or humidity, the corresponding scene mutation rate can be calculated, thereby constructing a set of samples for the model to learn.

[0042] The power prediction sample is obtained by calculating the new energy power generation prediction results at each time point in the historical period, which reflects the estimated future power based on meteorological conditions and load characteristics at that time.

[0043] Furthermore, identify an error sample set between the power prediction value and the actual power value of the historical period samples. The error sample set refers to the difference between the predicted value and the actual power for each period, which is used to measure the accuracy of the model. The error sample set can be positive or negative, representing over-prediction or under-prediction respectively.

[0044] Then, use the vector with the scene mutation rate sample and the power prediction sample as inputs, that is, describe the characteristics of each historical scene through the degree of meteorological mutation and the predicted power value, which is used to train or fit the model, and further predict the internal relationship between the error and the input variables.

[0045] Use the corresponding error sample set as a label identifying the influence relationship for linear fitting. Linear fitting is a statistical method used to establish a linear relationship between input variables and output variables. By minimizing the prediction error, a function model is trained so that when the scene mutation rate and the predicted power are known, the magnitude of the error can be estimated, thereby forming a fitting curve representing the relationship between the two. Finally, obtain the influence relationship curve, which reflects the variation law between the scene mutation rate and the power prediction error. The influence of weather mutation on the prediction accuracy can be quantified through the influence relationship curve.

[0046] Furthermore, this application also includes: analyzing the scene mutation rate according to the influence relationship curve to obtain a power prediction influence index; wherein, the power prediction influence index is used to measure the influence of extreme weather scenarios on the short-term power prediction accuracy, and the power prediction influence index and the scene mutation rate are in a direct proportional relationship.

[0047] Specifically, analyzing the scene mutation rate according to the influence relationship curve, inputting the mutation rate of the current or historical period, so as to obtain the corresponding prediction error trend. The influence relationship curve reflects the quantitative relationship between the severity of meteorological condition changes and the prediction error. By inputting the current mutation rate value, the possible influence of this weather change on the power prediction accuracy can be deduced, providing a basis for response when extreme weather comes. The power prediction influence index is the result obtained by analyzing the influence relationship curve, used to quantify the potential error caused by the weather mutation in the current period to the short-term power prediction result, and is a numerical index for evaluating the prediction risk.

[0048] Based on the risk identification mechanism, a mechanism for dynamically adjusting the time phase window length according to the prediction risk, i.e., the prediction influence index, can be designed. The time phase window length refers to the time range of the data referenced forward or backward by the model when making a prediction. The purpose of adjusting the window length is to use a longer time period to obtain a more robust trend judgment in the case of low risk, while shortening the time window in the case of high risk to respond more quickly to drastic changes, so as to more accurately track the short-term power fluctuations brought about by meteorological mutations.

[0049] The power prediction influence index is used to measure the influence of extreme weather scenarios on the short-term power prediction accuracy. The higher the power prediction influence index, the stronger the prediction uncertainty, and the larger the error space, and vice versa.

[0050] Furthermore, this application also includes: the short-term power prediction window regulator is set with a multi-stage phase expansion ratio, wherein the multi-stage phase expansion ratio corresponds to a multi-stage power prediction influence interval; the short-term power prediction window regulator obtains the phase expansion ratio of the corresponding stage according to the power prediction influence interval to which the power prediction influence index belongs, wherein the phase expansion ratio is less than 1.

[0051] Specifically, the short-term power prediction window regulator is set with a multi-stage phase expansion ratio to divide different prediction risk levels. The phase expansion ratio is a parameter for measuring the scaling ratio of the time window length. The multi-stage setting can adopt different window adjustment strategies according to different levels of risk. For example, when the risk is small, the phase expansion ratio is close to 1, indicating that only a slight adjustment of the time window is required, while when the risk is extremely high, the expansion ratio is much less than 1, and the window is greatly compressed.

[0052] The multi-stage phase expansion ratio corresponds to the multi-stage power prediction influence interval, indicating that each phase expansion ratio corresponds one-to-one with a specific influence index interval. The power prediction influence interval divides the influence index into multiple levels according to the numerical range. For example, when the influence index is between 0 and 30, it is a low-risk interval; between 30 and 70 is a medium risk; and exceeding 70 is a high-risk interval. Each interval is associated with a corresponding expansion ratio to guide how the window length changes.

[0053] Subsequently, the short-term power prediction window regulator obtains the phase expansion ratio of the corresponding stage according to the power prediction influence interval to which the power prediction influence index belongs, indicating that the regulator will automatically determine which interval the current risk index belongs to and match the preset time window adjustment scheme for the interval. The prediction model can automatically select the most appropriate reference time range for different weather change situations, making the prediction more flexible and accurate.

[0054] A phase expansion ratio less than 1 indicates that all adjustment actions are based on the window contraction strategy. Because in extreme weather or when there are large mutations, referring to a longer time window may introduce outdated or invalid data, which instead affects the prediction accuracy. Shortening the window, that is, the phase expansion ratio is less than 1, helps to focus on the recent data fluctuations and improve the model's response ability to rapid changes.

[0055] Furthermore, this application also includes: recording the previous time phase corresponding to the previous time period; expanding the previous time phase according to the phase expansion ratio to obtain an expanded time phase, and setting the expanded time phase in the time period under the extreme weather scenario until the extreme weather detection network detects that the current time period switches to the next extreme weather scenario and re-performs phase expansion or phase recovery.

[0056] Specifically, record the previous time phase corresponding to the previous time period. The time phase refers to the length of the historical time window referenced by the power prediction model, which determines the time range of the data used by the model in each prediction. For example, if the current phase is 10 minutes, it means that the data of the most recent 10 minutes is used for each prediction. To ensure the coherence and traceability of the prediction behavior, it is necessary to judge whether the time phase needs to be adjusted or restored in the future.

[0057] Next, expanding the previous time phase according to the phase expansion ratio means scaling the original time phase according to the previously obtained phase expansion ratio to obtain a new time window. An expansion ratio less than 1 indicates that the phase has been compressed, that is, the time range of the historical data used by the model has been shortened.

[0058] After that, set the extended time phase during the period in the extreme weather scenario, indicating that the new compressed time phase only takes effect within the period when it is detected that the current situation belongs to the extreme weather. This can prevent overly frequent predictions in normal weather, avoid resource waste, and also ensure sensitivity in special meteorological states.

[0059] Finally, until the extreme weather detection network detects that the current period switches to the next extreme weather scenario and re - performs phase extension or phase recovery, which means continuously monitoring the weather conditions. Once the detection network finds that the current period is no longer in the original extreme weather situation and may have changed to normal or entered a new extreme weather state, it will re - evaluate the new phase extension ratio, perform a new time phase adjustment, or restore the prediction phase to the default value. This can ensure that the prediction model not only has high adaptability but also maintains overall prediction stability.

[0060] Furthermore, this application also includes: outputting the phase extension ratio corresponding to the power prediction impact index, and the calculation formula includes: ; where is the phase extension ratio, is the adjustment parameter for controlling the prediction sensitivity, is the power prediction impact index.

[0061] Specifically, outputting the phase extension ratio corresponding to the power prediction impact index means calculating a corresponding time - phase scaling ratio according to the current power prediction impact index, which is used to adjust the frequency of short - term power prediction, achieve dynamic adjustment in the power prediction model, and make it adaptively change the prediction frequency according to the degree of drastic changes in weather or environment, so as to improve the model's response ability to mutations.

[0062] is the phase extension ratio, which is a proportionality factor for adjusting the power prediction time interval. The phase extension ratio is usually less than 1, meaning that in a mutation environment, the prediction time window will be shortened, making the model more sensitive. is the adjustment parameter for controlling the prediction sensitivity, which is a set constant used to control the sensitivity of the prediction to risk changes. The larger the adjustment parameter, the more likely it is to respond to changes in the power prediction impact index. is the power prediction impact index, which is a quantitative risk index. The larger the value of the power prediction impact index, the stronger the impact of the current extreme weather on power prediction, and the higher - frequency prediction behavior is required.

[0063] Furthermore, this application also includes: if the extreme weather detection network detects that the current period switches to the next extreme weather scenario, update the phase extension ratio and re - perform phase extension according to the updated phase extension ratio; if the extreme weather detection network detects that the current period returns to a non - extreme weather scenario, perform phase recovery on the extended time phase.

[0064] Specifically, the extreme weather detection network is continuously used to monitor the current meteorological state. Once it is recognized that the current period is no longer in the previous extreme weather scenario but has entered a new extreme weather scenario, such as changing from heavy rainfall to strong wind or from high temperature to low temperature, the original prediction conditions are no longer applicable. At this time, the power prediction impact index will be re-evaluated, and a new phase expansion ratio will be calculated accordingly. The updated expansion ratio will be used to re-adjust the time window length used by the prediction model, that is, phase expansion, so that the extreme weather detection network can adapt to the new sudden scenario in a timely manner and continuously maintain high sensitivity and accuracy of the prediction.

[0065] The extreme weather detection network has a mechanism for dynamically shrinking the prediction frequency. When the extreme weather detection network determines that the current weather condition has returned to the normal state, such as stable wind speed, stopped precipitation, and temperature returning to a reasonable range, it is considered that there is no longer a need to maintain a high-frequency prediction mode. At this time, the prediction time window shortened due to sudden weather before will be restored to the default or basic phase, such as changing from predicting every 5 minutes before to predicting every 10 minutes, which can avoid the phenomenon of overuse of resources or excessive prediction noise in the stable scenario, thus ensuring the model operation efficiency and prediction stability.

[0066] In summary, the short-term power prediction method for large-scale new energy under extreme weather provided by this application has the following technical effects: By achieving the technical goal of dynamically adjusting the power prediction time phase window based on extreme weather detection and scenario mutation rate analysis, the technical effects of improving the response sensitivity and fitting accuracy of short-term power prediction under extreme weather and ensuring the stable operation and dispatching efficiency of the power grid are achieved.

[0067] Embodiment 2, based on the same inventive concept as the short-term power prediction method for large-scale new energy under extreme weather in the foregoing embodiment, this application also provides a short-term power prediction device for large-scale new energy under extreme weather. Please refer to the appendix Figure 2, including: a judgment module 11, configured to construct an extreme weather detection network, determine whether the current period is an extreme weather scenario according to the extreme weather detection network, and calculate a scenario mutation rate if the current period is an extreme weather scenario, where the scenario mutation rate is the mutation rate of the extreme weather scenario in the current period and the corresponding scenario in the previous period; an analysis module 12, configured to establish an influence relationship curve between the scenario mutation rate sample and the power prediction sample, analyze the scenario mutation rate according to the influence relationship curve, and obtain a power prediction influence index; an output module 13, configured to input the power prediction influence index into a short-term power prediction window regulator, and output a phase expansion ratio corresponding to the power prediction influence index; a prediction module 14, configured to perform phase expansion on the period in the extreme weather scenario according to the phase expansion ratio to obtain an extended extended time phase, call a short-term power prediction model to perform power prediction according to the extended time phase, and output a power prediction result.

[0068] Further, the large-scale new energy short-term power prediction device under extreme weather is further configured to: collect the meteorological feature dimensions of the current period; the extreme weather detection network includes a convolutional layer, a fully connected layer, and an activation output layer, where the convolutional layer is configured to perform local change feature convolution on the meteorological feature dimensions, the fully connected layer is configured to perform feature fusion on the convolutional meteorological feature dimensions, and the activation output layer is configured to output a result indicating whether the current period is an extreme weather scenario according to the fusion features.

[0069] Further, the large-scale new energy short-term power prediction device under extreme weather is further configured to: if the current period is an extreme weather scenario, the expression for calculating the scenario mutation rate includes: ; where is the scenario mutation rate of the current period, is the number of meteorological feature dimensions, is the value of the i-th meteorological feature at the current period t, is the value of the i-th meteorological feature in the previous period t-1, is a very small positive number used to prevent the denominator from being zero.

[0070] Further, the large-scale new energy short-term power prediction device under extreme weather is further configured to: where the scenario mutation rate sample is obtained by calculating the scenario mutation rate of historical period samples, and the power prediction sample is obtained by calculating the power prediction values of the historical period samples; identify the error sample set between the power prediction value and the actual power value of the historical period samples; perform linear fitting with the scenario mutation rate sample and the power prediction sample as input vectors and the corresponding error sample set as the label indicating the influence relationship to obtain the influence relationship curve.

[0071] Further, the large-scale new energy short-term power prediction device under extreme weather is also used for: analyzing the scene mutation rate according to the influence relationship curve to obtain a power prediction influence index; wherein, the power prediction influence index is used to measure the influence of the extreme weather scene on the short-term power prediction accuracy rate, and the power prediction influence index and the scene mutation rate are in a direct proportional relationship.

[0072] Further, the large-scale new energy short-term power prediction device under extreme weather is also used for: the short-term power prediction window regulator is set with a multi-stage phase expansion ratio, wherein the multi-stage phase expansion ratio corresponds to a multi-stage power prediction influence interval; the short-term power prediction window regulator obtains the phase expansion ratio of the corresponding stage according to the power prediction influence interval to which the power prediction influence index belongs, wherein the phase expansion ratio is less than 1.

[0073] Further, the large-scale new energy short-term power prediction device under extreme weather is also used for: recording the previous time phase corresponding to the previous time period; expanding the previous time phase according to the phase expansion ratio to obtain an expanded time phase, and setting the expanded time phase in the time period under the extreme weather scene until the extreme weather detection network detects that the current time period switches to the next extreme weather scene to re-expand the phase or restore the phase.

[0074] Further, the large-scale new energy short-term power prediction device under extreme weather is also used for: outputting the phase expansion ratio corresponding to the power prediction influence index, and the calculation formula includes: ; wherein, is the phase expansion ratio, is the adjustment parameter for controlling the prediction sensitivity, is the power prediction influence index.

[0075] Further, the large-scale new energy short-term power prediction device under extreme weather is also used for: if the extreme weather detection network detects that the current time period switches to the next extreme weather scene, updating the phase expansion ratio and re-expanding the phase according to the updated phase expansion ratio; if the extreme weather detection network detects that the current time period returns to the non-extreme weather scene, restoring the phase of the expanded time phase.

[0076] In this specification, the various embodiments are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The large-scale new energy short-term power prediction method and specific examples under extreme weather in the foregoing Embodiment 1 are equally applicable to the large-scale new energy short-term power prediction device under extreme weather in this embodiment. Through the foregoing detailed description of the large-scale new energy short-term power prediction method under extreme weather, those skilled in the art can clearly understand the large-scale new energy short-term power prediction device under extreme weather in this embodiment. Therefore, for the sake of simplicity of the specification, it will not be elaborated herein.

[0077] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0078] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is also intended to include these changes and modifications.

Claims

1. A large-scale renewable energy short-term power forecasting method under extreme weather conditions, characterized in that: Methods include: Construct an extreme weather detection network, and determine whether the current period is an extreme weather scene based on the extreme weather detection network. If the current period is an extreme weather scene, calculate the scene mutation rate, which is the mutation rate between the extreme weather scene in the current period and the corresponding scene in the previous period; Establishing an influence relationship curve between scene mutation rate samples and power prediction samples, analyzing the scene mutation rate according to the influence relationship curve, and obtaining a power prediction influence index; Inputting the power prediction impact index into a short-term power prediction window regulator, and outputting a phase expansion ratio corresponding to the power prediction impact index; According to the phase expansion comparison, the phase expansion is performed for the time period under the extreme weather scenario to obtain the extended time phase after the expansion, and the short-term power prediction model is called to perform power prediction according to the extended time phase, and the power prediction result is output.

2. The method according to claim 1, characterized in that The method for determining whether the current period is an extreme weather scene according to the extreme weather detection network includes: Collect the meteorological characteristic dimensions of the current period; The extreme weather detection network includes a convolution layer, a fully connected layer and an activated output layer, wherein the convolution layer is used to perform local change feature convolution on the meteorological feature dimension, the fully connected layer is used to perform feature fusion on the convolved meteorological feature dimension, and the activated output layer is used to output the result of whether the current time period is an extreme weather scene based on the fused features.

3. The method according to claim 2, characterized in that If the current period is an extreme weather scenario, the expression for calculating the scene mutation rate includes: ； in, is the scene mutation rate of the current period, is the number of meteorological feature dimensions, is the value of the ith meteorological feature in the current period t, is the value of the ith meteorological feature in the previous period t-1, is a very small positive number used to prevent the denominator from being zero.

4. The method according to claim 1, characterized in that Establish the influence relationship curve between scene mutation rate samples and power prediction samples. include: The scene mutation rate samples are obtained by calculating the scene mutation rate of the historical period samples, and the power prediction samples are obtained by calculating the power prediction values ​​of the historical period samples; Identify a set of error samples between the power prediction value and the actual power value of the historical period sample; The scene mutation rate samples and the power prediction samples are used as input vectors, and a corresponding error sample set is used as a label to identify the influence relationship to perform linear fitting to obtain the influence relationship curve.

5. The method according to claim 1, characterized in that Analyze the scene mutation rate according to the impact relationship curve to obtain a power prediction impact index; The power forecast impact index is used to measure the impact of extreme weather scenarios on the accuracy of short-term power forecasts, and the power forecast impact index is in positive proportion to the scenario mutation rate.

6. The method according to claim 1, characterized in that Outputting a phase expansion ratio corresponding to the power prediction impact index, the method comprising: The short-term power prediction window regulator is provided with a multi-stage phase expansion ratio, wherein the multi-stage phase expansion ratio corresponds to a multi-stage power prediction influence interval; The short-term power prediction window regulator obtains a phase expansion ratio of a corresponding stage according to a power prediction influence interval to which the power prediction influence indicator belongs, wherein the phase expansion ratio is less than 1.

7. The method according to claim 6, characterized in that Phase expansion is performed for a period of time in an extreme weather scenario according to the phase expansion ratio, the method comprising: Record the last time phase corresponding to the last period; The previous time phase is expanded according to the phase expansion ratio to obtain an extended time phase, and the extended time phase is set in a time period under an extreme weather scenario until the extreme weather detection network detects that the current time period switches to the next extreme weather scenario and re-expands the phase or recovers the phase.

8. The method according to claim 1, characterized in that The phase expansion ratio corresponding to the power prediction impact index is output, and the calculation formula includes: ； in, is the phase expansion ratio, To control the prediction sensitivity, It is the power prediction impact indicator.

9. The method according to claim 6, characterized in that Until the extreme weather detection network detects that the current period switches to the next weather scene and re-performs phase expansion or phase recovery, the method includes: If the extreme weather detection network detects that the current period switches to the next extreme weather scenario, the phase expansion ratio is updated, and the phase expansion is performed again according to the updated phase expansion ratio; If the extreme weather detection network detects that the current period returns to a non-extreme weather scenario, the extended time phase is phase recovered.

10. A large-scale new energy short-term power prediction device under extreme weather conditions, characterized in that: The steps for implementing the large-scale new energy short-term power prediction method under extreme weather conditions as described in any one of claims 1 to 9 include: A judgment module is used to construct an extreme weather detection network, and judge whether the current period is an extreme weather scene according to the extreme weather detection network. If the current period is an extreme weather scene, the scene mutation rate is calculated, and the scene mutation rate is the mutation rate of the extreme weather scene in the current period and the corresponding scene in the previous period; An analysis module, used to establish an influence relationship curve between scene mutation rate samples and power prediction samples, and analyze the scene mutation rate according to the influence relationship curve to obtain a power prediction influence index; An output module, used for inputting the power prediction impact index into a short-term power prediction window regulator, and outputting a phase expansion ratio corresponding to the power prediction impact index; The prediction module is used to perform phase expansion for the time period under the extreme weather scenario according to the phase expansion comparison, obtain the extended time phase after expansion, call the short-term power prediction model to perform power prediction according to the extended time phase, and output the power prediction result.

Images