Power load analysis method, apparatus, device, medium, and product

Abstract

The application relates to a power load analysis method, device, equipment, medium and product. The method comprises the following steps: performing mode recognition on historical load data corresponding to a to-be-analyzed power system and prediction event information at a target analysis time, so as to obtain a power load mode of the to-be-analyzed power system at the target analysis time; and analyzing the power load of the to-be-analyzed power system at the target analysis time according to the power load mode, the historical load data, the prediction event information and a weather forecast sequence corresponding to the to-be-analyzed power system, so as to obtain a load analysis result. The method can accurately predict the power load in a data imbalance scene.

Description

Technical Field

[0001] This application relates to the field of smart grid and power system load forecasting technology, and in particular to a power load analysis method, apparatus, equipment, medium and product. Background Technology

[0002] With the acceleration of urbanization and the advancement of new power system construction, the power grid loads of megacities such as Shenzhen and Shanghai are characterized by large total load, significant peak-to-valley differences, and high sensitivity to external disturbances. Especially under the influence of localized, high-frequency, and non-periodic events such as typhoon warnings, large-scale exhibitions, and holiday lockdowns, electricity consumption behavior exhibits strong abrupt changes and multimodal distributions, making traditional load forecasting methods difficult to adapt.

[0003] Traditional techniques rely on unsupervised clustering to classify day types and then train heavy-duty deep models, or use a single global model for end-to-end prediction.

[0004] However, traditional technologies suffer from problems such as lack of event guidance in pattern recognition, rigid model architecture, and limited output formats. This leads to niche, high-impact patterns being overlooked in data-imbalanced scenarios, making it impossible to accurately predict power load in such scenarios. Summary of the Invention

[0005] Therefore, it is necessary to provide a power load analysis method, device, equipment, medium, and product that can accurately predict power load in scenarios with unbalanced data, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for analyzing power load, including:

[0007] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0008] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0009] In one embodiment, based on power load patterns, historical load data, predicted event information, and the weather forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain load analysis results, including:

[0010] Determine the load analysis model corresponding to the power load pattern;

[0011] Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time, and the load forecast demand in the load analysis results is obtained.

[0012] Based on the load forecasting requirements, the power risk of the power system to be analyzed under the target analysis time is assessed to obtain the forecast credibility in the load analysis results;

[0013] Based on the prediction reliability, load dispatch instructions are generated from the load analysis results.

[0014] In one embodiment, the power risk of the power system to be analyzed at the target analysis time is assessed based on the load forecasting demand to obtain the forecast reliability in the load analysis results, including:

[0015] A forecast risk assessment is performed on at least one load forecast value included in the load forecast demand to obtain the forecast confidence level.

[0016] In one embodiment, load scheduling instructions from the load analysis results are generated based on the prediction confidence level, including:

[0017] If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0018] If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0019] In one embodiment, pattern recognition is performed based on historical load data corresponding to the power system to be analyzed and predicted event information at the target analysis time to obtain the power load pattern of the power system to be analyzed at the target analysis time, including:

[0020] Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0021] The event weights corresponding to the predicted event information are determined based on the sensitivity coefficient and the event duration.

[0022] Pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0023] In one embodiment, pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time, including:

[0024] Obtain power load data corresponding to at least one preset load mode;

[0025] A similarity assessment is performed based on the event weight, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0026] The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

[0027] Secondly, this application also provides a power load analysis device, comprising:

[0028] The identification module is used to perform pattern recognition based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time, so as to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0029] The analysis module is used to analyze the power load of the power system under the target analysis time based on the power load pattern, historical load data, predicted event information and the meteorological forecast sequence corresponding to the power system to be analyzed, and obtain the load analysis results.

[0030] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0031] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0032] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0033] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0034] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0035] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0036] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0037] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0038] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0039] The aforementioned power load analysis methods, devices, equipment, media, and products identify the power load pattern corresponding to the target analysis time by combining historical load data of the power system under analysis with predicted event information at the target analysis time. Then, guided by the power load pattern, they integrate historical load data, predicted event information, and weather forecast sequences to perform power load analysis, obtaining the load analysis results for the target analysis time. This approach improves the accuracy of pattern segmentation in multimodal load scenarios and enhances the adaptability and reliability of load analysis under different load patterns, enabling accurate power load prediction even in data-imbalanced scenarios. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a diagram illustrating the application environment of the power load analysis method in one embodiment;

[0042] Figure 2 This is a flowchart illustrating a power load analysis method in one embodiment;

[0043] Figure 3 This is a flowchart illustrating the power load analysis method in yet another embodiment;

[0044] Figure 4This is a flowchart illustrating the power load analysis method in another embodiment;

[0045] Figure 5 This is a structural block diagram of a power load analysis device in one embodiment;

[0046] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0049] The power load analysis method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 sends a power load analysis request to server 104. Server 104 receives the power load analysis request, executes the power load analysis method, and feeds back the load analysis results to terminal 102. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0050] In one exemplary embodiment, such as Figure 2As shown, a power load analysis method is provided, which can be applied to... Figure 1 Taking the server in the example of this, the explanation includes:

[0051] Step 201: Based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time, perform pattern recognition to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0052] The power system to be analyzed can be a high-density urban distribution network area; historical load data can be the daily-scale load time series data of the power system to be analyzed that has been collected in the past, for example, with a sampling interval of 15 minutes (i.e., a total of 96 sampling points in a single day), the active power value at each moment is recorded to form a set of historical daily load curves; the target analysis time can be the target future period for load analysis or prediction, for example, the predicted future 1-24 hours; the predicted event information can be the information on external disturbance events that can be obtained in advance under the target analysis time, including but not limited to meteorological, social activities and public management events; the power load pattern can be a load category with similar electricity consumption characteristics and event-driven patterns obtained by clustering historical load and event data, such as summer air conditioning peak type, typhoon emergency low-valley type and large-scale exhibition high-load type.

[0053] Optionally, based on the predicted event information, the historical load data corresponding to the power system to be analyzed is compared with the power load data corresponding to the pre-set load pattern, so as to determine the load pattern corresponding to the power load data with the highest similarity as the power load pattern under the target analysis time.

[0054] Step 202: Based on the power load pattern, historical load data, predicted event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0055] Among them, the meteorological forecast sequence can be the meteorological time series forecast data of the corresponding area of ​​the power system to be analyzed within the target analysis time range, including meteorological indicators such as temperature and humidity, forming a continuous sequence with a fixed duration, used to reflect the impact of meteorological factors on electricity load; the load analysis results include, but are not limited to, load forecast demand, forecast reliability, and load dispatch instructions.

[0056] Optionally, historical load sequences, weather forecast sequences, and predicted event information are vectorized and concatenated to form a multi-source fusion feature vector in a unified format. Load analysis models corresponding to power load patterns are retrieved from a pre-built model library. During the model inference phase, the Dropout mechanism is kept continuously enabled, and multiple forward propagations are performed using Monte Carlo sampling to obtain multiple sets of different raw load forecast values. Statistical calculations are performed on these multiple sets of raw forecast values ​​to sequentially solve for load forecast demand, forecast confidence, and load dispatch instructions, thus obtaining the final load analysis results.

[0057] The aforementioned power load analysis method identifies the power load pattern corresponding to the target analysis time by combining historical load data of the power system under analysis with predicted event information at the target analysis time. Then, guided by the power load pattern, it integrates historical load data, predicted event information, and weather forecast sequences to perform power load analysis, obtaining the load analysis results for the target analysis time. This approach improves the accuracy of pattern segmentation in multimodal load scenarios and enhances the adaptability and reliability of load analysis under different load patterns, enabling accurate power load prediction even in data-imbalanced scenarios.

[0058] In one exemplary embodiment, such as Figure 3 As shown, the analysis of the power load of the power system under the target analysis time is performed based on the power load pattern, the historical load data, the predicted event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, to obtain the load analysis results, including:

[0059] Step 301: Determine the load analysis model corresponding to the power load pattern.

[0060] The load analysis model can be a lightweight GRU (Gated Recurrent Unit) sub-model, which is a neural network model used to complete the calculation, analysis and prediction of power load.

[0061] Optionally, in this embodiment, for each type of load pattern, historical load, event, and meteorological sample data corresponding to that pattern are extracted; a single-layer lightweight GRU network structure is uniformly adopted to train a dedicated sub-model for each type of load pattern; after training, a one-to-one mapping relationship between load patterns and lightweight GRU sub-models is established; when power load analysis is required, the determined power load pattern is obtained; then, according to the preset pattern-model mapping relationship, a search is performed in the model library; finally, the lightweight GRU sub-model corresponding to the power load pattern is selected, i.e., the load analysis model.

[0062] Step 302: Based on the load analysis model, the power load of the power system to be analyzed at the target analysis time is analyzed according to historical load data, predicted event information and weather forecast sequence to obtain the load forecast demand in the load analysis results.

[0063] Among them, the load forecast demand can be the predicted value of the power load under the target analysis time.

[0064] Optionally, historical load data, forecast event information, and weather forecast sequences are concatenated into vectors to generate a multi-source fusion feature vector with a uniform format and regular dimensions, which serves as the standard input to the load analysis model. The concatenated feature vector is then input into the currently selected lightweight GRU sub-model. A random dropout mechanism is continuously enabled throughout the model's inference and prediction phase, meaning the random discard logic of neurons is not disabled. Based on the Monte Carlo method, multiple rounds of forward propagation calculations are repeatedly executed. Each round, due to the random dropout mechanism, yields a set of independent raw load forecast values, generating multiple sets of discrete load forecast results, i.e., load forecast requirements.

[0065] Step 303: Assess the power risk of the power system to be analyzed under the target analysis time based on the load forecasting requirements, and obtain the forecast credibility in the load analysis results.

[0066] Among them, prediction confidence can be an inverse measure of the reliability level of the prediction result, which can be determined by the prediction uncertainty width U. t To represent and implement. And U t The smaller the value, the narrower the prediction interval and the higher the prediction reliability; U t The larger the value, the greater the uncertainty and the lower the reliability of the prediction.

[0067] Optionally, the uncertainty width obtained by Monte Carlo random deactivation can be extracted from the load forecast demand, and then the uncertainty width can be used to characterize the forecast reliability.

[0068] Step 304: Based on the prediction confidence level, generate load scheduling instructions from the load analysis results.

[0069] Among them, the load dispatching instruction can be an executable power grid operation instruction formulated for the power system to be analyzed, including generator output arrangement and spinning reserve capacity configuration, etc., and the parameters contained in the load dispatching instruction can be dynamically adjusted according to the prediction reliability to guide the actual dispatching work of the power grid under the target analysis time.

[0070] Optionally, a pre-set threshold for predictive reliability is obtained, dividing the reliability into three levels: high, medium, and low. Each level corresponds to a scheduling strategy rule, which specifies the configuration standards for unit output and spinning reserve capacity under different levels. The predictive reliability is compared with the preset threshold to determine the reliability level to which the predictive reliability belongs, thereby determining the corresponding generator output arrangement and spinning reserve capacity. Finally, load scheduling instructions are generated from the load analysis results based on the generator output arrangement and spinning reserve capacity.

[0071] In this embodiment, by combining a lightweight model with multi-source data fusion analysis, the shortcomings of traditional methods in terms of poor adaptability to multimodal and strongly disturbed load scenarios and insufficient prediction accuracy are alleviated. This enables the system to adapt to complex operating conditions of high-density urban power grids affected by external factors such as extreme weather, large-scale public events, and control policies.

[0072] In an exemplary embodiment, assessing the power risk of the power system to be analyzed at the target analysis time based on the load forecasting requirements to obtain the forecast credibility in the load analysis results includes: assessing the forecasting risk of at least one load forecast value included in the load forecasting requirements to obtain the forecast credibility.

[0073] The load forecast values ​​can be multiple discrete load forecast results output by the model.

[0074] Optionally, from the generated load forecast demand, all load forecast values ​​to be analyzed are extracted; based on all load forecast values ​​to be analyzed, statistical indicators such as standard deviation, 95% confidence interval, and uncertainty width are calculated to measure the dispersion and fluctuation range of multiple load forecast values; among them, the more dispersed the load forecast values ​​and the larger the uncertainty width, the higher the probability that the actual load deviates from the forecast result, and the higher the corresponding forecast risk level; conversely, the forecast risk is lower; finally, the quantified forecast risk is converted into a corresponding numerical or level forecast credibility.

[0075] In this embodiment, by using multiple sets of load forecast values ​​to conduct quantitative assessment of forecast risk and generate forecast credibility, the reliability of load forecast results under different scenarios can be objectively reflected. At the same time, risk indicators are transformed into standardized credibility parameters, which improves the practicality and reliability of power load analysis under complex disturbance conditions.

[0076] In an exemplary embodiment, generating a load scheduling instruction from the load analysis results based on the prediction confidence includes: if the prediction confidence is greater than a first confidence threshold and the prediction confidence is not greater than a second confidence threshold, then generating a load scheduling instruction based on the prediction confidence and the first confidence threshold; if the prediction confidence is greater than the second confidence threshold, then generating a load scheduling instruction based on the prediction confidence, the first confidence threshold, and the second confidence threshold.

[0077] The first confidence threshold can be a preset confidence level threshold, used to define the starting boundary of the medium confidence level range; the second confidence threshold can be a preset confidence level threshold, used as the dividing boundary of the high confidence level range; the first confidence threshold is less than the second confidence threshold.

[0078] Optionally, a pre-set first confidence threshold and a second confidence threshold are obtained. The predicted confidence is compared with the two thresholds. If the predicted confidence is greater than the first confidence threshold but not greater than the second confidence threshold, it is determined to be in the medium confidence interval. The rules corresponding to this interval are called, and parameters such as the output and reserve capacity of the power grid generators are determined by combining the current predicted confidence with the first confidence threshold, so as to generate the corresponding load dispatch instructions.

[0079] If the prediction confidence level is determined to be greater than the second confidence level threshold, it is classified as a high confidence level interval. The rules corresponding to the high confidence level interval are called, and parameters such as the power output and reserve capacity of the power grid generator units are determined based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold, and corresponding load dispatch instructions are generated.

[0080] Optionally, if the prediction confidence is not greater than the first confidence threshold, it is determined to be in the low confidence interval, and thus the incremental spinning reserve capacity is determined to be 0, that is, the power system does not need to increase the reserve capacity under the target analysis time.

[0081] In this embodiment, a two-level confidence threshold is used to classify the prediction confidence into intervals, and load dispatch instructions are generated differently for each level. This allows the power grid to increase reserves when the prediction confidence is moderate and ensure safety when the prediction confidence is extremely low, thereby significantly improving the overall operational stability while ensuring power supply reliability.

[0082] In an exemplary embodiment, pattern recognition is performed based on historical load data corresponding to the power system to be analyzed and predicted event information at the target analysis time to obtain the power load pattern of the power system to be analyzed at the target analysis time. This includes: determining the sensitivity coefficient and event duration corresponding to the predicted event information; determining the event weight corresponding to the predicted event information based on the sensitivity coefficient and event duration; and performing pattern recognition based on the event weight and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0083] The sensitivity coefficient is used to characterize the inherent impact of different types and intensities of external events on the power load. Different events (such as typhoons, large-scale exhibitions, and extreme high temperatures) correspond to different coefficients, and the stronger the event's disturbance to the load, the larger the sensitivity coefficient value. The event duration can be the complete length of time from the start to the end of the predicted event, that is, the effective period during which the event disturbs the regional load. The event weight is used to distinguish the importance of different time periods in the load similarity calculation, and the event's effective period is given a higher weight, while the weight of ordinary time periods is relatively low.

[0084] Optionally, based on the specific type and intensity level of the predicted event information, the sensitivity coefficient matching the event is retrieved from a preset parameter library; simultaneously, the event duration is calculated based on the start and end times recorded in the predicted event information. Then, according to preset weighting rules, the sensitivity coefficient and event duration are fused to obtain the event weight corresponding to each load sampling time; the event weight is combined with the historical load data of the power system to be analyzed, and a weighted Euclidean distance formula is constructed using the event weight as a coefficient. The similarity of the load curves is calculated using the weighted Euclidean distance formula, and finally, the power load pattern corresponding to the target analysis time is determined.

[0085] In this embodiment, by combining the event sensitivity coefficient and the duration to dynamically calculate the event weight, differentiated weight configuration for different events and different periods of action is achieved. This effectively strengthens the disturbance characteristics of external events on the load, alleviates the defects of traditional pattern recognition where event-related load characteristics are diluted and special load patterns are easily misjudged, and thus improves the accuracy and reliability of power load pattern recognition under multimodal and strong disturbance conditions.

[0086] In an exemplary embodiment, pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time. This includes: acquiring power load data corresponding to at least one preset load pattern; performing similarity assessment based on event weights, historical load data corresponding to the power system to be analyzed, and power load data corresponding to at least one preset load pattern to determine the similarity assessment value between each preset load pattern and the power system to be analyzed; and selecting the preset load pattern with the largest similarity assessment value among the preset load patterns as the power load pattern of the power system to be analyzed at the target analysis time.

[0087] Among them, the preset load modes include, but are not limited to, summer air conditioning peak mode, post-holiday resumption of work ramp-up mode, typhoon emergency low-load mode, and large-scale exhibition high-load mode; the similarity assessment value can be the quantitative result of the similarity assessment output, and the larger the value of the similarity assessment value, the closer the shape and change pattern of the two load curves are.

[0088] Optionally, all candidate preset load patterns are retrieved from a pre-built load pattern library, and virtual baseline load data corresponding to each preset load pattern is extracted one by one. Then, the event weight is used as the weighting coefficient for each sampling time, and combined with the historical load data of the power system to be analyzed and the virtual baseline load data of each preset load pattern, a weighted Euclidean distance algorithm is used to perform similarity calculations to obtain the similarity evaluation value for each preset load pattern. The similarity evaluation values ​​of all candidate preset load patterns are numerically compared, and the preset load pattern with the highest similarity evaluation value is selected as the power load pattern corresponding to the power system to be analyzed at the target analysis time.

[0089] In this embodiment, similarity quantification is achieved by combining virtual benchmark load data based on preset load patterns with event weights and weighted Euclidean distance. This effectively strengthens the load characteristics corresponding to external events, alleviates the shortcomings of traditional Euclidean distance similarity calculation that easily dilutes key features and has low accuracy in special pattern recognition, and thus improves the accuracy of power load pattern recognition.

[0090] In one exemplary embodiment, such as Figure 4 As shown, the power load analysis method can also include: the first step is pattern recognition, which uses a weighted clustering mechanism guided by local events to accurately classify multimodal and unbalanced load patterns, providing clean and highly interpretable input for subsequent forecasting; the second step is lightweight forecasting, which constructs a dedicated single-layer GRU lightweight quantum model (i.e., load analysis model) based on the identified load patterns (i.e., power load patterns) to achieve high-precision, low-latency short-term load point forecasting, and realizes real-time perception of pattern drift and dynamic model switching through a dual-condition criterion; the third step is adaptive and scheduling closed-loop, which maintains long-term forecast accuracy through model rolling fine-tuning, combines probabilistic forecasting to quantify uncertainty, and directly maps forecast risk to spinning reserve capacity.

[0091] Optionally, this embodiment addresses the challenge of load identification in high-load-density urban power grids under complex environments by constructing a local event-guided load pattern recognition mechanism for high-density urban power grids. High-load-density cities experience multiple overlapping effects from high-frequency meteorological disturbances, large-scale exhibition economies, and routine traffic control, resulting in highly non-stationary, strongly abrupt, and multimodal coupled characteristics of their power grid load. In this scenario, traditional unsupervised clustering methods rely solely on historical statistical similarity, easily leading to pattern confusion between "normal high-temperature days" and "high-energy-consuming exhibition days," or the failure of key load patterns due to the "few extreme event samples" being overwhelmed by the "majority of normal samples."

[0092] This embodiment unifies, classifies, and aligns typical high-frequency disturbance events in high-load-density cities with time data, forming a structured event label system that strictly matches load data. This addresses the problems of fragmented event information and the inability to quantify and integrate it into the clustering process. The multi-dimensional event label vector for day i is constructed as follows:

[0093]

[0094] in, This is the structured event label vector corresponding to day i, which corresponds one-to-one with the load sampling time. ; This represents the total number of event types, including meteorological, social activity, and public management events. is the hierarchical quantization code value for the i-th day and the k-th type of event, and the value represents the intensity of the event's impact; ⊤ is the vector transpose symbol to ensure the alignment of the input dimensions.

[0095] Typical event categories in this embodiment include: meteorological events: typhoon warning levels are divided into four levels: blue, yellow, orange, and red; social events: large-scale events such as the China Hi-Tech Fair, the China Cultural Expo, and concerts; and public management events: special holidays or situations such as lockdowns during the Spring Festival and areas under epidemic control.

[0096] Single-type event hierarchical quantization coding formula:

[0097]

[0098] in, This represents the level of the k-th type of event on day i, with 0 for no event and a larger value for higher levels. is the impact weighting coefficient for the k-th type of event, used to distinguish the degree of disturbance to the load caused by different events.

[0099] For high-impact events such as red typhoons, the impact period is extended by 24 hours to form a persistent impact marker, expressed as:

[0100]

[0101] in, Encode the event component at the t-th sampling time on the i-th day; The start time of a high-impact event is 24 hours; 24 hours is the forced extension duration of a high-impact event to ensure that subsequent load anomalies are fully captured.

[0102] This embodiment defines the 1-norm of the event vector to determine whether a day is a day of significant disturbance:

[0103]

[0104] in, Let be the total disturbance intensity of the event on day i; when When a major event is identified, the pattern library is automatically re-identified and updated.

[0105] To highlight the impact of event-sensitive periods, this embodiment designs an event-driven period weight (i.e., event weight) generation function:

[0106]

[0107] in, : This represents the sensitivity coefficient (i.e., sensitivity index) for the k-th type of event, ranging from 0.5 to 2.5. It is calibrated through historical data analysis, specifically using a grid search combined with profile coefficient optimization, ultimately determining the threshold for a red typhoon warning. Large-scale exhibitions 1.5, ordinary holidays take 0.5; Let be the set of time periods (i.e., the duration of events) that affect the k-th type of event on day i. For indicator functions, The value is 1 when the condition is met, and 0 otherwise. This method automatically amplifies the difference weights during active event periods and suppresses noise interference during non-critical periods when ensuring the clustering algorithm calculates similarity.

[0108] This embodiment uses time-based weighting and a weighted Euclidean distance to measure the load curves of any two days. and The similarity between the two methods replaces the traditional Euclidean distance, solving the problem of the failure of the traditional distance under event perturbation.

[0109]

[0110] in, The weighted similarity distance between the load curves of day i and day j is given; the smaller the distance, the closer the load patterns are. T is the number of sampling points in a day; T=96 corresponds to 24 hours with a 15-minute sampling interval. Let be the load value for day i in time period t; Let be the load value for day j in time period t.

[0111] To improve the semantic consistency and matching speed of the patterns, this embodiment constructs a virtual baseline load curve for each type of load pattern m. The calculation method is as follows:

[0112]

[0113] in, Let be the virtual baseline load value for the m-th load pattern at time t; The set of all sample days for the current attribution pattern m; Let m be the number of samples contained in the m-th pattern. ; For set The load value of the i-th sample at time t.

[0114] Typical load patterns in this embodiment (i.e., preset load patterns) include, but are not limited to, peak summer air conditioning load, post-holiday work resumption ramp-up load, typhoon emergency low load load, and high load load for large-scale exhibitions. Virtual baseline curves are stored in the pattern library for rapid matching of new samples.

[0115] This embodiment uses newly acquired daily load curves. The optimal pattern assignment is determined by calculating a weighted distance with all virtual baseline curves in the pattern library.

[0116]

[0117] in, The optimal mode number to which the new sample is finally assigned (i.e., the mode number corresponding to the power load mode). It is the set of all recognized patterns in the current pattern library; The dynamic time period weights corresponding to the new sample days; Let be the load value of the new sample at time t; This is the virtual baseline curve for pattern m.

[0118] When the system detects a significant local event (such as...) If the existing pattern cannot be matched for 3 consecutive days, the clustering re-identification process will be automatically triggered. Based on the load and event data of the most recent 7 to 14 days, weighted clustering will be re-executed, and the pattern library and virtual baseline curve will be dynamically updated to ensure that the pattern library is always adapted to the latest load changes.

[0119] Optionally, this embodiment, based on accurate pattern segmentation, constructs an independent lightweight GRU sub-model for each pattern category to address the problems of traditional heavy models being unable to be deployed at the edge, insufficient global model accuracy, and lag in static switching response. Simultaneously, a dual-condition dynamic switching mechanism is designed to ensure prediction robustness. Specifically:

[0120] After classifying typical load patterns, this embodiment configures a dedicated lightweight prediction sub-model for each type of pattern m to avoid performance degradation of a single global model in multimodal scenarios. The sub-model inputs fuse three types of information: historical load (i.e., historical load data), weather forecasts (i.e., weather forecast sequences), and event tags (i.e., prediction event information), fully covering both endogenous load patterns and exogenous disturbances, and is constructed as follows:

[0121]

[0122] in, Input the feature vector into the model at time t; The historical load sequence for the previous p time periods, p=8 (corresponding to 32 sampling points), is used to capture the short-term time series trend of the load; This is a temperature and humidity forecast sequence for the next q hours, where q = 24 hours, used to reflect the driving effect of meteorology on the load; Encode the events for day t to indicate whether events such as typhoons, exhibitions, or lockdowns occur on that day. This is the vector transpose symbol, ensuring input dimension alignment.

[0123] In this embodiment, the sub-model adopts a single-layer gated recurrent unit (GRU) construction pattern to construct a dedicated sub-model, and the number of parameters is controlled within a certain range. Within this range, it meets the low latency and low memory requirements of edge terminals. GRU gated update calculation:

[0124]

[0125]

[0126]

[0127]

[0128] in: To update the gate, control the proportion of historical information retained; To reset the door, control the degree to which historical information is ignored; This is the candidate hidden state; Output the hidden state at the current moment; sigmoid is the activation function; tanh is the hyperbolic tangent activation function; Multiply matrices element by element; , , , , , Here is the weight matrix of the GRU network; , , This is a bias term.

[0129] Load point prediction output formula:

[0130]

[0131] in, Let m be the final load forecast value of the m-th sub-model at time t+1; The current hidden state of the GRU; Output the layer weight vector for the m-th sub-model; The bias is set for the output layer of the m-th sub-model; This is a vector transpose. Each sub-model has the same structure but independent parameters, achieving precise modeling specific to each mode and avoiding accuracy degradation under multimodal loads.

[0132] This embodiment designs a dynamic switching criterion based on continuous deviation to address load pattern drift caused by sudden events in actual operation. Let... This represents the predicted value of the current activated sub-model at time t. For the actual load value, the normalized relative error is defined as:

[0133]

[0134] in, The normalized relative error at time t; The actual load value at time t; Let be the load forecast value of the m-th sub-model at time t; This represents the typical daily maximum load under the m-mode, used to eliminate dimensional differences between different modes.

[0135] The formula for calculating typical daily maximum load is as follows:

[0136]

[0137] in, The total number of sample days included in the m-th pattern; Let be the set of all historical sample days belonging to the m-th load pattern; i is the i-th sample day in the m-th pattern; t is the t-th load sampling time within the day, T=96; It represents the maximum load value in the load curve for the entire day of the i-th sample day.

[0138] To avoid noise-induced false triggering and frequent model oscillations, this embodiment designs a switching mechanism that combines continuous error exceeding limits with a cooling period as two conditions:

[0139]

[0140]

[0141] in: This is a mode switching trigger flag; 1 indicates triggering, and 0 indicates no triggering. The preset normalized relative error threshold, ; This is the last time the mode was switched. The cooling-off period is fixed at 24 hours to prevent repeated switching within a short period of time.

[0142] When the error exceeds the limit for two consecutive time points and the cooling-off period is met, the system automatically restarts pattern recognition and switches sub-models; if the switching fails for three consecutive times, it automatically reverts to the hybrid mode sub-model trained on full data to ensure uninterrupted system operation.

[0143] Optionally, to address the issue of power load forecasting models degrading over time due to changes in user electricity consumption habits, the commissioning of new commercial complexes, and the continuous increase in electric vehicle penetration, as well as the dynamic evolution of the external environment, this embodiment introduces a weekly rolling fine-tuning mechanism and a probabilistic forecasting output structure. This promotes a leap from traditional static point forecasting to an intelligent closed-loop system that integrates rolling cognition, risk quantification, and scheduling coordination. Specifically:

[0144] Every Sunday morning (e.g., 2:00 AM), during the typical off-peak load period, the update process is automatically initiated. Using the latest accumulated weekly actual load data and corresponding meteorological observation data, online incremental learning and fine-tuning optimization are performed on the currently active lightweight forecast sub-model.

[0145]

[0146] in, Fine-tuning the loss function for the m-th sub-model; The total number of samples to be fine-tuned is M = 96 × 7 = 672; The actual load of the j-th sample; These are the model's predicted values; These are the updated model parameters; These are the model parameters before the update. Here are the EWC regularization coefficients, λ= ; It uses the L2 norm. The entire fine-tuning process is completed locally on the edge device, without uploading the raw payload or event data to the central server. This reduces communication bandwidth requirements while fully protecting user data privacy and system security.

[0147] This embodiment uses Monte Carlo Dropout to quantify uncertainty, upgrading point prediction to a probability distribution output:

[0148]

[0149]

[0150] in, This is the forecast mean (final point forecast value), which is the mean value corresponding to the load forecast demand; For the predicted standard deviation; The number of samples is N=100; The 95% confidence interval is half-width; This represents the 95% confidence quantile of a normal distribution. This embodiment defines the prediction uncertainty width. This directly reflects the reliability of the prediction.

[0151] This embodiment establishes a direct mapping from predicted risk to scheduling instructions, enabling risk-driven scheduling:

[0152]

[0153] in, The additional spinning reserve capacity (MW) required at the current moment is the spinning reserve capacity increment. The low-risk threshold (i.e., the first confidence threshold) is set. ; This is the high-risk threshold (i.e., the second confidence threshold). ; , For example, the sensitivity coefficient. , The above approach demonstrates the scheduling principle that the higher the risk, the stronger the backup response. That is, when uncertainty is low, economic operations are maintained; when a typhoon or event causes a sharp drop in forecast reliability, the backup margin is automatically increased.

[0154] This embodiment ultimately outputs three types of results that can be directly used for power grid dispatch: a deterministic point prediction of power load at a future time (t+1), serving as the core result of load forecasting; a 95% confidence interval and uncertainty width (i.e., forecast credibility) for load forecasting, enabling a quantitative expression of forecast risk; and a suggested value for the incremental spinning reserve capacity automatically generated based on forecast uncertainty, directly converting the forecast results into executable dispatch instructions (i.e., load dispatch instructions). These outputs collectively constitute a complete decision-making basis for load forecasting and safe, economical dispatching of high-density urban power grids.

[0155] To more comprehensively demonstrate this solution, this embodiment presents a power load analysis method, specifically including:

[0156] 1. Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0157] 2. Determine the event weights corresponding to the predicted event information based on the sensitivity coefficient and the event duration;

[0158] 3. Obtain power load data corresponding to at least one preset load mode;

[0159] 4. Based on the event weights, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, perform a similarity assessment to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0160] 5. The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed under the target analysis time.

[0161] 6. Determine the load analysis model corresponding to the power load pattern;

[0162] 7. Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time to obtain the load forecast demand in the load analysis results.

[0163] 8. Conduct a forecast risk assessment on at least one load forecast value included in the load forecast demand to obtain the forecast credibility;

[0164] 9. If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0165] 10. If the prediction confidence level is greater than the second confidence level threshold, then generate a load scheduling instruction based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0166] The specific process of the above steps can be found in the description of the above method embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0167] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0168] Based on the same inventive concept, this application also provides a power load analysis device for implementing the power load analysis method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power load analysis device embodiments provided below can be found in the limitations of the power load analysis method described above, and will not be repeated here.

[0169] In one exemplary embodiment, such as Figure 5 As shown, a power load analysis device is provided, including: an identification module 51 and an analysis module 52, wherein:

[0170] The identification module 51 is used to perform pattern recognition based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time, so as to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0171] Analysis module 52 is used to analyze the power load of the power system under the target analysis time based on the power load pattern, historical load data, predicted event information and the meteorological forecast sequence corresponding to the power system to be analyzed, and obtain the load analysis results.

[0172] In one embodiment, the analysis module 52 is further configured to:

[0173] Determine the load analysis model corresponding to the power load pattern;

[0174] Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time, and the load forecast demand in the load analysis results is obtained.

[0175] Based on the load forecasting requirements, the power risk of the power system to be analyzed under the target analysis time is assessed to obtain the forecast credibility in the load analysis results;

[0176] Based on the prediction reliability, load dispatch instructions are generated from the load analysis results.

[0177] In one embodiment, the analysis module 52 is further configured to:

[0178] A forecast risk assessment is performed on at least one load forecast value included in the load forecast demand to obtain the forecast confidence level.

[0179] In one embodiment, the analysis module 52 is further configured to:

[0180] If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0181] If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0182] In one embodiment, the identification module 51 is further configured to:

[0183] Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0184] The event weights corresponding to the predicted event information are determined based on the sensitivity coefficient and the event duration.

[0185] Pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0186] In one embodiment, the identification module 51 is further configured to:

[0187] Obtain power load data corresponding to at least one preset load mode;

[0188] A similarity assessment is performed based on the event weight, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0189] The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

[0190] Each module in the aforementioned power load analysis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0191] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores load analysis results. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power load analysis method.

[0192] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0193] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0194] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0195] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0196] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0197] Determine the load analysis model corresponding to the power load pattern;

[0198] Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time, and the load forecast demand in the load analysis results is obtained.

[0199] Based on the load forecasting requirements, the power risk of the power system to be analyzed under the target analysis time is assessed to obtain the forecast credibility in the load analysis results;

[0200] Based on the prediction reliability, load dispatch instructions are generated from the load analysis results.

[0201] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0202] A forecast risk assessment is performed on at least one load forecast value included in the load forecast demand to obtain the forecast confidence level.

[0203] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0204] If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0205] If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0206] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0207] Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0208] The event weights corresponding to the predicted event information are determined based on the sensitivity coefficient and the event duration.

[0209] Pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0210] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0211] Obtain power load data corresponding to at least one preset load mode;

[0212] A similarity assessment is performed based on the event weight, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0213] The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

[0214] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0215] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0216] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0217] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0218] Determine the load analysis model corresponding to the power load pattern;

[0219] Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time, and the load forecast demand in the load analysis results is obtained.

[0220] Based on the load forecasting requirements, the power risk of the power system to be analyzed under the target analysis time is assessed to obtain the forecast credibility in the load analysis results;

[0221] Based on the prediction reliability, load dispatch instructions are generated from the load analysis results.

[0222] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0223] A forecast risk assessment is performed on at least one load forecast value included in the load forecast demand to obtain the forecast confidence level.

[0224] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0225] If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0226] If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0227] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0228] Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0229] The event weights corresponding to the predicted event information are determined based on the sensitivity coefficient and the event duration.

[0230] Pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0231] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0232] A similarity assessment is performed based on the event weight, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0233] The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

[0234] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0235] Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time.

[0236] Based on power load patterns, historical load data, forecast event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

[0237] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0238] Determine the load analysis model corresponding to the power load pattern;

[0239] Based on the load analysis model, and according to historical load data, forecast event information and weather forecast sequence, the power load of the power system to be analyzed is analyzed at the target analysis time, and the load forecast demand in the load analysis results is obtained.

[0240] Based on the load forecasting requirements, the power risk of the power system to be analyzed under the target analysis time is assessed to obtain the forecast credibility in the load analysis results;

[0241] Based on the prediction reliability, load dispatch instructions are generated from the load analysis results.

[0242] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0243] A forecast risk assessment is performed on at least one load forecast value included in the load forecast demand to obtain the forecast confidence level.

[0244] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0245] If the prediction confidence is greater than the first confidence threshold and the prediction confidence is not greater than the second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold.

[0246] If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

[0247] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0248] Determine the sensitivity coefficient and event duration corresponding to the predicted event information;

[0249] The event weights corresponding to the predicted event information are determined based on the sensitivity coefficient and the event duration.

[0250] Pattern recognition is performed based on event weights and historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

[0251] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0252] A similarity assessment is performed based on the event weight, the historical load data corresponding to the power system to be analyzed, and the power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed.

[0253] The preset load mode with the highest similarity evaluation value among all preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

[0254] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0255] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0256] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0257] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for analyzing power load, characterized in that, The method includes: Pattern recognition is performed based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time to obtain the power load pattern of the power system to be analyzed under the target analysis time. Based on the power load pattern, the historical load data, the predicted event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, the power load of the power system to be analyzed at the target analysis time is analyzed to obtain the load analysis results.

2. The method according to claim 1, characterized in that, The process involves analyzing the power load of the power system under the target analysis time based on the power load pattern, historical load data, predicted event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, to obtain load analysis results, including: Determine the load analysis model corresponding to the power load pattern; Based on the load analysis model, the power load of the power system to be analyzed at the target analysis time is analyzed according to the historical load data, the predicted event information and the weather forecast sequence, and the load forecast demand in the load analysis results is obtained. The power risk of the power system to be analyzed under the target analysis time is assessed based on the load forecast demand, and the forecast credibility in the load analysis results is obtained. Based on the prediction confidence level, load scheduling instructions are generated from the load analysis results.

3. The method according to claim 2, characterized in that, The step of assessing the power risk of the power system under analysis at the target analysis time based on the load forecast demand to obtain the forecast reliability in the load analysis results includes: A prediction risk assessment is performed on at least one load prediction value included in the load prediction demand to obtain the prediction confidence level.

4. The method according to claim 3, characterized in that, The step of generating load scheduling instructions from the load analysis results based on the prediction confidence level includes: If the prediction confidence is greater than a first confidence threshold and the prediction confidence is not greater than a second confidence threshold, then a load scheduling instruction is generated based on the prediction confidence and the first confidence threshold, wherein the first confidence threshold is less than the second confidence threshold. If the prediction confidence level is greater than the second confidence level threshold, then a load scheduling instruction is generated based on the prediction confidence level, the first confidence level threshold, and the second confidence level threshold.

5. The method according to any one of claims 1 to 4, characterized in that, The step of performing pattern recognition based on historical load data corresponding to the power system to be analyzed and predicted event information at the target analysis time to obtain the power load pattern of the power system to be analyzed at the target analysis time includes: Determine the sensitivity coefficient and event duration corresponding to the predicted event information; The event weight corresponding to the predicted event information is determined based on the sensitivity coefficient and the event duration. Pattern recognition is performed based on the event weights and the historical load data corresponding to the power system to be analyzed to determine the power load pattern of the power system to be analyzed at the target analysis time.

6. The method according to claim 5, characterized in that, The step of performing pattern recognition based on the event weights and historical load data corresponding to the power system to be analyzed, to determine the power load pattern of the power system to be analyzed at the target analysis time, includes: Obtain power load data corresponding to at least one preset load mode; A similarity assessment is performed based on the event weights, historical load data corresponding to the power system to be analyzed, and power load data corresponding to at least one preset load mode, to determine the similarity assessment value between each preset load mode and the power system to be analyzed. The preset load mode with the highest similarity evaluation value among the preset load modes is taken as the power load mode of the power system to be analyzed at the target analysis time.

7. A power load analysis device, characterized in that, The device includes: The identification module is used to perform pattern recognition based on the historical load data of the power system to be analyzed and the predicted event information under the target analysis time, so as to obtain the power load pattern of the power system to be analyzed under the target analysis time. The analysis module is used to analyze the power load of the power system to be analyzed at the target analysis time based on the power load pattern, the historical load data, the predicted event information, and the meteorological forecast sequence corresponding to the power system to be analyzed, and to obtain the load analysis results.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images