Instrumentation power management system for a substation

Abstract

This application relates to the field of power management, specifically disclosing an instrumentation power management system for substations. It first performs multi-dimensional component analysis on mixed raw load data based on business functions, finely distinguishing different types of load growth sources such as protection and monitoring. Then, utilizing the deep time-series analysis capabilities of a large model, it independently predicts the future evolution trend of each component, forming a clear future load structure diagram. Furthermore, by combining these predicted future load data with real-time updated battery characteristic parameters, it performs forward-looking risk indicator extrapolation, transforming abstract current values ​​into concrete and intuitive risk quantification indicators such as future backup time after an accident. In this way, the system can generate structured early warnings including risk attribution months or even longer in advance.

Description

Technical Field

[0001] This application relates to the field of power management, and more specifically, to an instrumentation power management system for substations. Background Technology

[0002] As a critical hub in the power system, the safe and stable operation of substations is essential to the entire power grid. Inside a substation, core instruments and meters for protection, measurement and control, and communication rely on a highly reliable DC power supply system for uninterrupted power. This ensures that in emergencies such as grid failures or AC power outages, these critical devices can still function normally and execute correct protection and control strategies. Therefore, effectively managing the health status of the substation instrument and meter power supply system to ensure its reliability and continuity is a fundamental requirement for the safe operation and maintenance of the power system.

[0003] However, existing power management solutions are generally inadequate in addressing incremental risks. Traditional management methods primarily focus on real-time monitoring and threshold alarms for basic parameters such as voltage and current, representing a reactive response mechanism either after the fact or during an incident. With the continuous construction and upgrading of smart substations, new instruments and equipment are being connected to the DC system in batches, leading to a slow and imperceptible increase in total load. This incremental load growth is like a frog being slowly boiled in water; a single increase in load may seem insignificant, but over time, it gradually erodes the safety margin initially designed into the power system. Existing management systems lack forward-looking assessment tools and cannot effectively identify this long-term trend, resulting in an information barrier between planning and actual operation. Often, problems are only discovered when the system load rate approaches the alarm threshold or when insufficient backup time is exposed during an incident. By then, power capacity expansion and upgrades are too late, posing a potentially significant threat to grid security. This lack of early warning due to the absence of forward-looking assessments of the matching between future loads and power supply capacity is a technical problem that urgently needs to be solved in the field of substation power management.

[0004] Therefore, an optimized instrumentation power management system for substations is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an instrumentation power management system for substations.

[0006] According to one aspect of this application, an instrumentation power management system for a substation is provided, comprising:

[0007] The multidimensional component parsing module is used to perform multidimensional component parsing on the original feeder data based on the feeder classification mapping table to obtain load time-series vectors for multiple device categories.

[0008] The battery characteristic parameter determination module is used to determine the current battery characteristic parameters based on real-time battery data and historical battery database. The current battery characteristic parameters include effective capacity and Pecker coefficient.

[0009] The multi-component time series analysis module is used to perform time series component prediction based on a large model on the load time series vectors of multiple device categories to obtain multi-component prediction profiles.

[0010] The risk indicator extrapolation module is used to extrapolate and calculate future key risk indicators based on multi-component prediction profiles and current battery characteristic parameters to obtain future key risk indicator profiles.

[0011] The early warning prompt determination module is used to determine whether to generate a structured early warning prompt based on the profile of future key risk indicators.

[0012] Compared to existing technologies, the instrumentation power management system for substations provided in this application first performs multi-dimensional component analysis on the mixed raw load data according to business functions, finely distinguishing different types of load growth sources such as protection and monitoring. Then, utilizing the deep time-series analysis capabilities of a large model, it independently predicts the future evolution trend of each component, forming a clear future load structure diagram. Furthermore, by combining these predicted future load data with real-time updated battery characteristic parameters, it performs forward-looking risk indicator extrapolation, transforming abstract current values ​​into concrete and intuitive risk quantification indicators such as future backup time in case of accidents. In this way, the system can generate structured early warnings containing risk attribution months or even longer in advance, revealing potential power supply and load mismatch problems at their nascent stage, thus achieving a fundamental shift from passive monitoring to proactive and accurate risk prediction. Attached Figure Description

[0013] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 This is a system block diagram of an instrumentation power management system for a substation according to an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of the data flow in an instrumentation power management system for a substation according to an embodiment of this application.

[0016] Figure 3 This is a block diagram of a multi-component timing analysis module in an instrumentation power management system for a substation according to an embodiment of this application.

[0017] Figure 4 This is a block diagram of a global load timing characteristic sensing unit in an instrumentation power management system for a substation according to an embodiment of this application. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0019] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0021] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0022] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0023] To address the specific technical problem of substations experiencing a gradual increase in total load due to continuous upgrades and renovations, which invisibly erodes the safety margin of the power system, while traditional monitoring methods lack forward-looking early warning capabilities, this application proposes an instrumentation power management system for substations. Specifically, this system, through in-depth analysis of raw feeder data, does not treat the total load in a general way, but rather, based on a pre-defined feeder classification mapping table, accurately decomposes the complex current data into multiple equipment category load time-series vectors with clear business meanings, such as protection, monitoring, and communication. Simultaneously, by analyzing real-time and historical battery data, the system dynamically determines the current effective capacity and key characteristic parameters of the batteries, such as the Pecker coefficient, establishing a precise physical model foundation for subsequent risk assessment. Based on this, a large model is used to perform in-depth prediction of the load time series of each analyzed equipment category, generating a multi-component prediction profile depicting the next few months or even a year, clearly revealing the independent growth trends of various load types. Furthermore, the system combines this future load forecast profile with established battery characteristic parameters to dynamically extrapolate key future risk indicators. This transforms abstract future current values ​​into concrete and perceptible risk indicators such as future charging module load rate and future backup time in case of an accident. Ultimately, by analyzing these extrapolated future risk indicator profiles, the system can determine whether a safety threshold will be exceeded at some point in the future. Based on this, it generates structured early warning prompts that include risk attribution, thereby transforming power management from passive state monitoring into a proactive and intelligent management paradigm that can anticipate, quantify, and support decision-making regarding risks.

[0024] The technical solution of this application proposes an instrument and meter power management system for substations. Figure 1 This is a system block diagram of an instrumentation power management system for a substation according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in an instrumentation power management system for a substation according to an embodiment of this application. Figure 1 and Figure 2As shown, the instrumentation power management system 100 for substations according to an embodiment of this application includes: a multi-dimensional component analysis module 110, used to perform multi-dimensional component analysis on the original feeder data based on a feeder classification mapping table to obtain load time-series vectors for multiple equipment categories; a battery characteristic parameter determination module 120, used to determine the current battery characteristic parameters based on real-time battery data and a historical battery database, the current battery characteristic parameters including effective capacity and Pecker coefficient; a multi-component time-series analysis module 130, used to perform time-series component prediction based on a large model on the load time-series vectors for multiple equipment categories to obtain a multi-component prediction profile; a risk indicator extrapolation module 140, used to extrapolate and calculate future key risk indicators based on the multi-component prediction profile and the current battery characteristic parameters to obtain a future key risk indicator profile; and an early warning prompt determination module 150, used to determine whether to generate a structured early warning prompt based on the future key risk indicator profile.

[0025] In the aforementioned instrumentation power management system 100 for substations, the multi-dimensional component analysis module 110 is used to perform multi-dimensional component analysis on the original feeder data based on a feeder classification mapping table to obtain load time-series vectors for multiple equipment categories. It should be understood that since the original feeder data is essentially a discrete, physical-level set of current measurements, it is unstructured in terms of business logic. Directly analyzing this raw data makes it difficult to reveal the intrinsic driving factors of load changes, and it is impossible to distinguish whether the overall change is caused by load growth in core protection equipment or power consumption fluctuations in auxiliary equipment. Therefore, in the technical solution of this application, multi-dimensional component analysis is performed on the original feeder data based on a feeder classification mapping table to obtain load time-series vectors for multiple equipment categories. This transforms the raw, undifferentiated physical measurement data into structured load component data with clear business attributes. This provides more information-rich input for subsequent trend prediction models, enabling them to perform targeted analysis of loads of different natures, thereby accurately locating the source of risk and improving the accuracy and interpretability of early warnings.

[0026] Specifically, in this embodiment, the multi-dimensional component analysis module is used to: acquire timestamped feeder snapshots; and perform load component aggregation calculations based on mapping on the feeder snapshots to be timestamped, based on a feeder classification mapping table, to obtain load time series data for multiple equipment categories. That is, the analysis process first acquires timestamped feeder snapshots, i.e., at a preset synchronization time point, the system collects the instantaneous current values ​​of all target DC feeders (e.g., feeders F01 to F05) within the substation and binds these current values ​​to a precise acquisition timestamp, forming a time-aligned raw data set. Next, the system performs mapping-based load component aggregation calculations. This step utilizes a pre-configured feeder classification mapping table, which defines the equipment category served by each feeder. For example, F01 and F02 serve protection equipment, F03 and F04 serve monitoring equipment, and F05 serves auxiliary equipment. The processing unit iterates through each feeder current in the snapshot, looks up its corresponding equipment category according to the mapping table, and arithmetically sums the feeder current values ​​belonging to the same category. For example, adding 12.1 amps of F01 and 13.0 amps of F02 yields a total current of 25.1 amps for the protection category loads; adding 18.5 amps of F03 and 17.0 amps of F04 yields a total current of 35.5 amps for the monitoring category loads. The result of this aggregation calculation, namely a load vector containing the same timestamp as the original snapshot and divided by equipment category, constitutes a data point in the load time series, laying the data foundation for subsequent multidimensional time series analysis.

[0027] In the aforementioned instrumentation power management system 100 for substations, the battery characteristic parameter determination module 120 is used to determine the current battery characteristic parameters based on real-time battery data and a historical battery database. These current battery characteristic parameters include effective capacity and the Peckert coefficient. It should be understood that since the performance parameters of a battery are not constant, its effective capacity will irreversibly decrease with factors such as service life, number of charge-discharge cycles, and ambient temperature. Furthermore, its actual discharge capacity at different discharge rates (i.e., the Peckert effect) also differs significantly from the ideal state. Therefore, in the technical solution of this application, the current battery characteristic parameters, including effective capacity and the Peckert coefficient, are further determined based on real-time battery data and a historical battery database to obtain a set of dynamic parameters that accurately characterize the battery pack's true health status and discharge characteristics at the current moment. This provides an accurate and reliable physical model benchmark for subsequent calculations of key future risk indicators, ensuring that the assessment results of core risk indicators such as future accident backup time are closer to actual operating conditions, thereby fundamentally improving the credibility and effectiveness of the entire early warning system.

[0028] More specifically, in a particular example of this application, the parameter determination process first involves periodically performing a battery health status assessment. The system retrieves the most recent complete charge-discharge process data from a historical battery database, calculates the actual amount of electricity released by the battery during that cycle using the ampere-hour integral method, and compares this with the battery's rated capacity to obtain the current state of health (SOH) percentage, for example, 90%. Subsequently, this SOH percentage is multiplied by the rated capacity (e.g., 500 ampere-hours) to calculate the current effective capacity as 450 ampere-hours. Next, the system performs Peckert coefficient calibration. It retrieves at least two records of stable discharge events with significantly different discharge currents from the historical database, for example, one discharge at 20 amperes for 20 hours and another discharge at 100 amperes for 3 hours. Using these data points, the Peckert coefficient, characterizing the nonlinear discharge characteristics of the battery pack, is calculated by solving the logarithmic form of the Peckert formula, for example, obtaining a coefficient value of 1.2. Ultimately, the dynamically determined effective capacity (450 Ah) and Pecker coefficient (1.2) are updated to the current battery characteristic parameters for use by the risk simulation module.

[0029] In the aforementioned instrumentation power management system 100 for substations, the multi-component time-series analysis module 130 is used to perform time-series component prediction based on a large model on the load time-series vectors of multiple equipment categories to obtain a multi-component prediction profile. It should be understood that, due to the different load evolution patterns of different equipment categories, their growth trends, periodicity, and responses to external factors vary significantly. Using a single model to predict the aggregated total load would mask these internal dynamics, resulting in insufficient prediction accuracy and an inability to perform effective risk attribution. Therefore, in the technical solution of this application, time-series component prediction based on a large model is further performed on the load time-series vectors of multiple equipment categories to obtain a multi-component prediction profile. This utilizes the powerful nonlinear pattern capture capability of the large model to perform deep time-series modeling and high-precision prediction for each independent business load component. This generates a refined multi-dimensional prediction view containing the future evolution trajectories of each component, not only improving the overall prediction accuracy but also providing a crucial, structured data foundation for subsequent risk quantification and attribution analysis.

[0030] Figure 3 This is a block diagram of a multi-component timing analysis module in an instrumentation power management system for a substation according to an embodiment of this application. Figure 3As shown in the embodiments of this application, the multi-component time-series analysis module 130 includes: a target device category load time-series data extraction unit 131, used to extract the load time-series vector of a first device category from the load time-series vectors of multiple device categories; a target device local load time-series feature extraction unit 132, used to extract local load time-series features from the load time-series vector of the first device category to obtain the sequence distribution of the local load time-series feature encoding vector of the first device category; a global load time-series feature perception unit 133, used to perform global load time-series feature context encoding on the sequence distribution of the local load time-series feature encoding vector of the first device category to obtain the global time-series feature encoding vector of the first device category; and a load prediction curve generation unit 134, used to perform feature decoding on the global time-series feature encoding vector of the first device category to obtain the load prediction curve of the first device category.

[0031] Specifically, the target device category load time-series data extraction unit 131 is used to extract the load time-series vector of the first device category from the load time-series vectors of multiple device categories. It should be understood that since the entire multi-component prediction module is designed using a divide-and-conquer strategy—that is, building a dedicated prediction model for each independent device category to capture its unique temporal dynamics—before initiating the prediction process for any specific category, the data for that specific category must be accurately separated from the set containing data for all categories. Therefore, in the technical solution of this application, the load time-series vector of the first device category is further extracted from the load time-series vectors of multiple device categories, thereby providing a clean and single data input source for subsequent feature extraction and prediction units. This ensures that the subsequent modeling process is fully focused on the inherent laws of the target category, avoiding cross-interference between different category data, which is a necessary prerequisite for achieving high-precision, highly targeted component prediction.

[0032] Specifically, the target device local load temporal feature extraction unit 132 is used to extract local load temporal features from the load temporal vector of the first device category to obtain the sequence distribution of the local load temporal feature encoding vector of the first device category. It should be understood that since the original load temporal vector of the first device category is a high-dimensional raw numerical sequence, it directly contains a large amount of redundant information and noise. If directly used for long-range dependency modeling, it will greatly increase the computational burden of subsequent models and may affect the robustness of predictions due to the inability to effectively distinguish between key patterns and irrelevant fluctuations. Therefore, in the technical solution of this application, local load temporal features are further extracted from the load temporal vector of the first device category to obtain the sequence distribution of the local load temporal feature encoding vector of the first device category, thereby transforming the original, unprocessed numerical sequence into a higher-level feature sequence composed of local temporal patterns. In this way, by effectively reducing the dimensionality and enhancing the features of the original data, a higher information density and easier-to-learn input can be provided to the subsequent global context awareness module, thereby significantly improving the efficiency and accuracy of the model in capturing long-range dependencies.

[0033] More specifically, in this embodiment, the target device local load temporal feature extraction unit is used to: input the load temporal vector of a first device category into a pre-trained one-dimensional convolutional neural network model to obtain the sequence distribution of the local load temporal feature encoding vector of the first device category. The one-dimensional convolutional neural network model contains multiple convolutional kernels, each of which has learned during training how to identify a specific local temporal pattern, such as intraday periodic fluctuations, weekend load troughs, or the initial pattern of seasonal load increases. During execution, these convolutional kernels perform sliding convolution operations on the input load temporal vector with a fixed stride. At each time step, a small segment of temporal data covered by the convolutional kernel is calculated with the kernel's weights to generate a value. The magnitude of this value characterizes the degree of matching between the current local segment and the specific pattern represented by the convolutional kernel. The calculation results of all convolutional kernels in the same step are combined into a vector, which is the local temporal feature encoding vector for that time step. As the convolution kernel slides across the entire input sequence, it generates a series of such encoded vectors in sequence. This ordered set of vectors ultimately constitutes the sequence distribution of the local temporal feature encoded vectors of the first device category load, and is passed to the next processing unit.

[0034] Specifically, the global load temporal feature sensing unit 133 is used to perform global load temporal feature context encoding on the sequence distribution of the local temporal feature encoding vector of the first device category load to obtain the global temporal feature encoding vector of the first device category. It should be understood that although the sequence distribution of the local temporal feature encoding vector of the first device category load generated in the previous stage contains rich pattern information, it is uniform in the time dimension and fails to distinguish which are key turning points indicating long-term trend changes and which are merely regular, periodic fluctuations. Therefore, in the technical solution of this application, the sequence distribution of the local temporal feature encoding vector of the first device category load is further subjected to global load temporal feature context encoding to obtain the global temporal feature encoding vector of the first device category. This transforms the processing paradigm of temporal information from passive aggregation to an active, discriminatory, and guided encoding, that is, before the final sequence encoding, an information theory-based, explicit temporal saliency preprocessing stage is introduced. This stage quantifies the uncertainty changes of the monitored load state, assigns an endogenous, dynamic importance score to each time step, and thus nonlinearly reshapes the entire temporal information flow. In this way, the key question of determining what constitutes a critical moment can be decoupled from the implicit learning task of the subsequent sequence encoder, allowing it to focus on deep modeling of the temporal dependencies between key events that have been identified as high-value events. This enables more accurate and efficient capture of the dynamic evolution logic of monitoring workloads.

[0035] Figure 4 This is a block diagram of the global load timing characteristic sensing unit in the instrumentation power management system for a substation according to an embodiment of this application. Figure 4 As shown in the embodiments of this application, the global load temporal feature sensing unit 133 includes: a load state sensing subunit 1331, used to perform a device category load state uncertainty measurement on each of the first device category load local temporal feature encoding vectors in the sequence distribution of the first device category load local temporal feature encoding vector to obtain the sequence distribution of the first device category load temporal entropy increment; a load temporal adjustment weight calculation subunit 1332, used to determine the sequence distribution of the first device category load temporal adjustment weight based on the sequence distribution of the first device category load temporal entropy increment; a weighted modulation subunit 1333, used to perform weight modulation on the sequence distribution of the first device category load local temporal feature encoding vector based on the sequence distribution of the first device category load temporal adjustment weight to obtain the sequence distribution of the adjusted first device category load vector; and a device category global temporal encoding subunit 1334, used to input the sequence distribution of the adjusted first device category load vector into a forward LSTM-based sequence encoder to obtain the first device category global temporal feature encoding vector.

[0036] Accordingly, the load state sensing subunit 1331 is used to perform a device category load state uncertainty measure on each of the first device category load local time series feature encoding vectors in the sequence distribution of the first device category load local time series feature encoding vectors to obtain the sequence distribution of the first device category load time series entropy increment, expressed by the following formula:

[0037]

[0038] Among them, v t It is the local temporal feature encoding vector of the first device category load at time node t in the sequence distribution of the local temporal feature encoding vector of the first device category load, v t,i It is v t The feature value at the i-th position in the vector, d is the feature value of v. t The length of p, where exp is the value of the logarithmic function to the base e. t,i It is v t The confidence level of the feature value at the i-th position, p t-1,i It is v t-1 The confidence level of the feature value at the i-th position, ΔE t v is the sequence distribution of the load entropy increment for the first device category. t and v t-1 The first device category load time sequence entropy increment.

[0039] It is understandable that, since the sequence distribution of the local temporal feature encoding vector of the first equipment category load does not explicitly distinguish which time points have regular and predictable state changes, and which are key events with higher information content that mark the inherent mutation of the load evolution pattern, the subsequent model may be interfered with by a large number of stable, low-information periods when learning long-term dependencies. Therefore, in the technical solution of this application, the uncertainty of the equipment category load state is further measured on each local temporal feature encoding vector of the first equipment category load in the sequence distribution of the first equipment category load to obtain the sequence distribution of the temporal entropy increment of the first equipment category load. This allows for the quantification of the node state uncertainty of each local feature vector in the sequence. The core of this approach is to borrow the concept of entropy from information theory, abstract the feature distribution of the monitored load at each moment into a measurable state determinism, and calculate the temporal entropy gain between adjacent time steps. In this way, it is possible to go beyond the direct dependence on the load feature values ​​themselves and instead focus on the inherent mutation of the feature evolution pattern, thereby introducing a higher-order meta-information into the system, namely the unexpectedness or information content of the event. This entropy increment sequence will serve as the basis for subsequent dynamic weight generation, enabling the model to autonomously identify and amplify the influence of moments that mark key phase transitions in the system.

[0040] Accordingly, the load timing adjustment weight calculation subunit 1332 is used to determine the sequence distribution of the load timing adjustment weight for the first equipment category based on the sequence distribution of the load timing entropy increment of the first equipment category, expressed by the following formula:

[0041]

[0042] Where exp is an exponential function with base e, β is a learnable scaling factor, and ΔE k Let α be the k-th time-series entropy increment of the first equipment category in the sequence distribution of the time-series entropy increments of the first equipment category, and T be the number of time-series entropy increments of the first equipment category in the sequence distribution of the time-series entropy increments of the first equipment category. t v in the sequence distribution of load timing adjustment weights for the first device category t The corresponding first device category load timing adjustment weight.

[0043] It is understandable that the temporal entropy increment sequence obtained in the previous step only characterizes the drastic change in load state at each time point, and it is not transformed into a normalized importance score that can be directly used to modulate the feature sequence, thus failing to directly achieve differentiated processing of information value at different times. Therefore, in the technical solution of this application, the sequence distribution of the temporal entropy increment of the first equipment category load is further determined based on the sequence distribution of the first equipment category load temporal adjustment weight, thereby constructing a dynamic, content-dependent weight generation mechanism. The principle is to map the unexpected index of entropy gain to an influence weight through a nonlinear function. A drastic state change, i.e., high entropy gain, will trigger a high weight, and vice versa. In this way, an endogenous attention mechanism can be achieved, allowing the model to autonomously and nonlinearly amplify the influence of those moments that mark the key phase transition of the monitoring load (such as the load step after a single equipment modification), while suppressing noise interference in normal, stable evolution periods (such as daily periodic fluctuations), thereby ensuring that the importance of the time dimension is no longer preset and uniformly decaying, but is completely determined by the dynamic evolution of the data itself.

[0044] Accordingly, the weighted modulation subunit 1333 is used to perform weighted modulation on the sequence distribution of the local temporal feature encoding vector of the first equipment category load based on the sequence distribution of the timing adjustment weight of the first equipment category load to obtain the adjusted sequence distribution of the first equipment category load vector, expressed by the following formula:

[0045] h t =v t ·α t

[0046] Among them, h tLet t be the t-th adjusted first device category load vector in the sequence distribution of the adjusted first device category load vector.

[0047] It is understandable that, since the sequence distribution of the load temporal adjustment weights for the first equipment category and the sequence distribution of the local temporal feature encoding vector for the first equipment category load are still two independent data streams, the calculated importance score has not yet been actually applied to the feature data itself. Therefore, the downstream encoder still faces unfiltered raw feature information. Therefore, in the technical solution of this application, based on the sequence distribution of the load temporal adjustment weights for the first equipment category, the sequence distribution of the local temporal feature encoding vector for the first equipment category load is further weighted to obtain the adjusted sequence distribution of the first equipment category load vector. This implements an information gating mechanism, directly applying the importance weights calculated in the previous step to the original monitoring-type load feature data stream. This substantially reshapes the time series input to the downstream encoder, ensuring that the load features of those critical, high-entropy gain moments are fully preserved or even amplified, while the influence of those unremarkable moments is weakened. This generates an adjusted sequence that is no longer a uniform time record, but one that has undergone importance preprocessing and dynamic information density changes, highlighting the most valuable historical inflection points for the final global feature encoding.

[0048] Accordingly, the device category global temporal coding subunit 1334 is used to input the sequence distribution of the adjusted first device category load vector into a forward LSTM-based sequence encoder to obtain the first device category global temporal feature coding vector, expressed by the following formula:

[0049] v z =LSTM{h1,h2,...,h n}

[0050] Among them, h1, h2, h n These represent the 1st, 2nd, and nth adjusted first-device-category load vectors in the sequence distribution of the adjusted first-device-category load vectors, respectively. LSTM is a sequence encoder based on forward LSTM. z This is the global temporal feature encoding vector for the first device category.

[0051] It is understandable that although the weighted sequence highlights key historical inflection points, it remains a feature sequence distributed along the timeline and has not yet been condensed into a fixed-dimensional global feature representation that comprehensively summarizes the entire historical evolution logic. This global representation is a necessary input for generating future prediction curves. Therefore, in the technical solution of this application, the sequence distribution of the adjusted first device category load vector is further input into a forward LSTM sequence encoder to obtain a global temporal feature encoding vector for the first device category. This leverages the powerful ability of forward long short-term memory networks to capture long-range temporal dependencies, operating on a sequence whose key information points have been pre-identified by upstream modules. This allows for in-depth modeling of the temporal correlations and evolutionary logic between historical events that have proven to be crucial. In this way, the entire historical sequence of monitoring loads, reshaped by nonlinear information, can be ultimately encoded into a highly condensed global temporal feature encoding vector that profoundly reflects its dynamic evolutionary essence, providing an optimal and information-rich feature input for subsequent load prediction curve generation.

[0052] In the aforementioned instrumentation power management system 100 for substations, the risk indicator extrapolation module 140 is used to perform future key risk indicator extrapolation calculations based on multi-component prediction profiles and current battery characteristic parameters to obtain future key risk indicator profiles. It should be understood that since the multi-component prediction profile and dynamically determined battery characteristic parameters are two independent and abstract datasets, they cannot directly reveal potential future operational risks. For example, a future load forecast value itself cannot inform maintenance personnel whether the power system will be overloaded or whether the backup time meets the procedural requirements. Therefore, in the technical solution of this application, future key risk indicator extrapolation calculations are further performed based on multi-component prediction profiles and current battery characteristic parameters to obtain future key risk indicator profiles. This deeply integrates the prediction of abstract data for the future with the assessment of the current equipment status, and through extrapolation using physical models and operating rules, transforms future load values ​​into perceptible, concrete risk indicators directly related to safety regulations. In this way, potential and long-term risks can be presented in a quantitative and intuitive way, enabling managers to see the evolution trend of system margin months or even years in advance, thereby achieving a fundamental shift in management model from passive response to proactive prevention.

[0053] Specifically, in this embodiment, the risk indicator extrapolation module includes: a load prediction curve extraction unit, used to extract the prediction curve of the total load from the multi-component prediction profile; and a risk indicator profile analysis unit, used to calculate the future charging module load rate and future backup time of the predicted total load at each time point in the total load prediction curve to obtain the future key risk indicator profile. More specifically, in a specific example of this application, the extrapolation calculation process is first performed by the load prediction curve extraction unit, which extracts the load prediction curves of all equipment categories (e.g., protection, monitoring, auxiliary) from the multi-component prediction profile, and performs arithmetic summation on these component load values ​​at each future time point (e.g., the 1st month, 2nd month...12th month) to generate a prediction curve representing the total DC load of the entire station. Subsequently, the risk indicator profile analysis unit traverses each time point on this total load prediction curve. For a future point in time, such as a predicted total load of 95 amps in 6 months, the unit first calculates the future charging module load rate by dividing the predicted total load value (95 amps) by the total rated output current of the substation charging modules (e.g., 100 amps), resulting in a future charging module load rate of 95%. Next, the unit calculates the future backup time under a hypothetical accident. It uses the same predicted total load value (95 amps) as the discharge current after a hypothetical accident and calls upon the currently determined battery characteristic parameters, namely the effective capacity (450 Ah) and the Peckert coefficient (1.2), applying the Peckert formula to calculate and deduce the backup time that the current battery pack can provide under this future load level. This calculation process is repeated at each point in time on the total load prediction curve, ultimately generating a sequence containing the charging module load rate and backup time for each future point in time. This sequence constitutes the aforementioned profile of future key risk indicators.

[0054] In the aforementioned instrumentation and power management system 100 for substations, the early warning determination module 150 is used to determine whether to generate a structured early warning based on a profile of future key risk indicators. It should be understood that since the profile of future key risk indicators derived in the previous step is merely a series of raw, uninterpreted quantitative data, it cannot automatically be converted into alarm information that provides direct guidance to maintenance personnel. Without an automated judgment mechanism, manual interpretation and comparison are still required, making true intelligent early warning impossible. Therefore, in the technical solution of this application, a structured early warning is further determined based on the profile of future key risk indicators, thereby establishing an automated decision-making logic layer. This logic layer continuously and proactively compares and judges the risk prediction value at each future time point with the preset multi-level safety procedure thresholds in real time. In this way, the analysis results of the entire system can be transformed from simple data presentation into actionable early warning events with clear guidance, truly realizing a closed loop from risk prediction to risk alarm, providing forward-looking, accurate, and automated decision support for the safe operation and maintenance of substations.

[0055] Specifically, in a concrete example of this application, the early warning generation process first involves the system loading judgment criteria from a pre-set risk threshold rule base. This rule base explicitly defines the alarm levels for different risk indicators; for example, a future charging module load rate exceeding 85% is a Level 1 warning, and exceeding 95% is a Level 2 warning; a future accident backup time of less than 2 hours is a Level 1 warning, and less than 1.5 hours is a Level 2 warning. Subsequently, the system iterates through each time point in the profile of future key risk indicators generated in the previous step. When processing the data point for the 8th month in the future, the system reads the predicted values ​​for that point as: a charging module load rate of 86% and an accident backup time of 1.9 hours. The system compares these two predicted values ​​with the thresholds in the rule base. Since 86% exceeds the Level 1 warning threshold of 85%, and 1.9 hours is less than the Level 1 warning threshold of 2 hours, the system determines that the warning conditions have been triggered. Based on this, the system generates a structured early warning message, which includes: a warning timestamp (8 months in the future), a warning level (Level 1 warning), a warning type (charging module load rate exceeding limits, insufficient backup time), a predicted value (86%, 1.9 hours), and the corresponding threshold (85%, 2 hours). This message is then pushed to the upper-level monitoring system interface to provide early risk warnings to maintenance personnel in a clear and specific manner. If no indicator is found to have reached the threshold after traversing the entire profile, the judgment process ends, and no warning is generated.

[0056] In summary, the instrumentation power management system for substations according to the embodiments of this application is explained. First, it performs multi-dimensional component analysis on the mixed raw load data based on business functions, finely distinguishing different types of load growth sources such as protection and monitoring. Then, utilizing the deep time-series analysis capabilities of a large model, it independently predicts the future evolution trend of each component, forming a clear future load structure diagram. Furthermore, by combining these predicted future load data with real-time updated battery characteristic parameters, it performs forward-looking risk indicator extrapolation, transforming abstract current values ​​into concrete and intuitive risk quantification indicators such as future backup time after an accident. In this way, the system can generate structured early warnings containing risk attribution months or even longer in advance, revealing potential power supply and load mismatch problems at their nascent stage, thereby achieving a fundamental shift from passive monitoring to proactive and accurate risk prediction.

[0057] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An instrument and meter power management system for substations, characterized in that, include: The multidimensional component parsing module is used to perform multidimensional component parsing on the original feeder data based on the feeder classification mapping table to obtain load time-series vectors for multiple device categories. The battery characteristic parameter determination module is used to determine the current battery characteristic parameters based on real-time battery data and historical battery database. The current battery characteristic parameters include effective capacity and Pecker coefficient. The multi-component time series analysis module is used to perform time series component prediction based on a large model on the load time series vectors of multiple device categories to obtain multi-component prediction profiles. The risk indicator extrapolation module is used to extrapolate and calculate future key risk indicators based on multi-component prediction profiles and current battery characteristic parameters to obtain future key risk indicator profiles. The early warning prompt determination module is used to determine whether to generate a structured early warning prompt based on the profile of future key risk indicators. The multi-component time series analysis module includes: The target device category load time series data extraction unit is used to extract the load time series vector of the first device category from the load time series vectors of multiple device categories; The target device local load temporal feature extraction unit is used to extract local load temporal features from the load temporal vector of the first device category to obtain the sequence distribution of the local load temporal feature encoding vector of the first device category; A global load timing feature sensing unit is used to perform global load timing feature context encoding on the sequence distribution of the local timing feature encoding vector of the first device category load to obtain the global timing feature encoding vector of the first device category. The load prediction curve generation unit is used to perform feature decoding on the global temporal feature encoding vector of the first device category to obtain the load prediction curve of the first device category. The global load timing feature sensing unit includes: The load status sensing subunit is used to perform a device category load status uncertainty measurement on each of the first device category load local time-series feature encoding vectors in the sequence distribution of the first device category load local time-series feature encoding vectors to obtain the sequence distribution of the first device category load time-series entropy increment; The load timing adjustment weight calculation subunit is used to determine the sequence distribution of the load timing adjustment weight of the first equipment category based on the sequence distribution of the load timing entropy increment of the first equipment category; The weighted modulation subunit is used to perform weighted modulation on the sequence distribution of the local temporal feature encoding vector of the first equipment category load based on the sequence distribution of the timing adjustment weight of the first equipment category load to obtain the adjusted sequence distribution of the first equipment category load vector. The device category global temporal coding subunit is used to input the sequence distribution of the adjusted first device category load vector into a forward LSTM-based sequence encoder to obtain the first device category global temporal feature coding vector.

2. The instrument and meter power management system for substations according to claim 1, characterized in that, The multidimensional component analysis module is used for: Get a timestamped feeder snapshot; Based on the feeder classification mapping table, load component aggregation calculations based on mapping are performed on the feeder snapshots to be timestamped to obtain load time series data for multiple device categories.

3. The instrument and meter power management system for substations according to claim 2, characterized in that, The target device local load temporal feature extraction unit is used to: input the load temporal vector of the first device category into a pre-trained one-dimensional convolutional neural network model to obtain the sequence distribution of the local load temporal feature encoding vector of the first device category.

4. The instrument and meter power management system for substations according to claim 1, characterized in that, The risk indicator derivation module includes: The load prediction curve extraction unit is used to extract the prediction curve of the total load from the multi-component prediction profile. The risk indicator profile analysis unit is used to calculate the future charging module load rate and future accident backup duration of the predicted total load at each time point in the prediction curve of the total load to obtain the future key risk indicator profile.

5. The instrument and meter power management system for substations according to claim 4, characterized in that, The risk indicator profile analysis unit is used to calculate the future charging module load rate and future accident backup time using the following formula: in, This represents the predicted total load at each time point in the total load prediction curve. and Define the total rated capacity and effective capacity of the charging module. For Peckert coefficients, For future charging module load rate, To provide backup time for future accidents.

Images