Method and system for generating safe power utilization decision

Abstract

The invention discloses a safe power utilization decision generation method and system, and the method comprises the steps: collecting operation parameters and environment parameters of power utilization equipment in real time through a sensor disposed in a power utilization network, integrating power grid dispatching data, user historical power utilization behavior data and meteorological data, and constructing a multi-source heterogeneous data pool; performing edge calculation processing on the data, including data cleaning, normalization and data fusion based on a dynamic space-time weight distribution algorithm, generating a dynamic power utilization state matrix, inputting the dynamic power utilization state matrix to a deep learning model, identifying an abnormal mode in equipment operation in real time, and quantifying a risk level based on a fuzzy logic algorithm, and based on the risk assessment result, constructing a multi-objective optimization model with safety, energy consumption cost and power supply continuity as optimization objectives, executing the optimized decision scheme through an Internet of Things controller, and monitoring the execution effect in real time.

Description

Technical Field

[0001] The present invention relates to the field of safe electricity use decision-making, and in particular to a method and system for generating safe electricity use decisions. Background Art

[0002] With the rapid development of smart grid and Internet of Things technologies, the safety and energy efficiency management of power systems face higher demands. Currently, power monitoring systems generally use fixed threshold alarm mechanisms, such as over-limit detection based on single sensor data (such as current or temperature). However, these approaches have significant drawbacks in complex power usage scenarios: First, they lack the ability to integrate heterogeneous data from multiple sources (such as device operating parameters, environmental conditions, and user behavior). Traditional static weight allocation algorithms struggle to adapt to sudden changes in device status or sensor failures, resulting in increased false alarm rates. Second, anomaly detection models often rely on single-source time series analysis, ignoring the topological relationships and spatial dependencies between devices. This results in low sensitivity for complex anomaly patterns such as arc faults and insulation degradation. Third, existing decision optimization methods often focus on a single objective (such as maximizing safety) and lack a dynamic balance between user preferences and constraints such as power continuity, making it difficult to generate actionable solutions. Fourth, they lack a closed-loop feedback mechanism after decision execution, making it difficult to distinguish between policy failure and external interference (such as sudden weather changes), leading to delayed model iteration.

[0003] To address the above issues, existing technologies have attempted to introduce data fusion algorithms and deep learning models for improvement, such as using weighted averaging to fuse multi-sensor data, or using convolutional neural networks to extract local features. However, such methods still have limitations: static fusion weights cannot reflect the real-time reliability differences of sensors, and rely on simple interpolation when data is missing, resulting in error accumulation; traditional neural networks have difficulty in modeling device topology relationships and time series dependencies at the same time, and lack the ability to quantify fuzzy uncertainties in risk assessment. In addition, multi-objective optimization algorithms often cannot meet real-time requirements due to their high computational complexity, and lack flexibility in emergency decision-making in extreme scenarios. Therefore, there is an urgent need for a safe electricity use decision-making method that takes into account accurate data perception, dynamic risk quantification, and multi-objective collaborative optimization to improve the level of safety protection and energy efficiency management in complex electricity use environments. Summary of the Invention

[0004] To achieve the above-mentioned and other related purposes, the present invention discloses a method for generating a safe electricity use decision, comprising the following steps:

[0005] Step 1: Sensors deployed in the power network collect real-time operating and environmental parameters of power equipment, integrate grid dispatch data, historical user power consumption data, and meteorological data, and build a multi-source heterogeneous data pool.

[0006] Performing edge computing processing on the data, including data cleaning, normalization, and data fusion based on a dynamic spatiotemporal weight allocation algorithm, to generate a dynamic power consumption state matrix;

[0007] The dynamic spatiotemporal weight allocation algorithm dynamically adjusts the fusion weight of each sensor based on the real-time correlation between sensor data and reference signals and the sensor stability score;

[0008] Step 2: Input the dynamic power consumption state matrix into a deep learning model to identify abnormal patterns in equipment operation in real time and quantify the risk level based on a fuzzy logic algorithm;

[0009] The deep learning model is a spatiotemporal graph convolution-attention hybrid network, which is used to capture device topology relationships and time series dependency features;

[0010] The risk assessment includes calculating the probability of failure and determining the risk level based on the current over-limit multiple and the temperature rise rate;

[0011] Step 3: Based on the risk assessment results, a multi-objective optimization model is constructed with safety, energy consumption cost, and power supply continuity as optimization objectives;

[0012] An improved non-dominated sorting genetic algorithm is used to globally optimize the control strategy, resource scheduling scheme, and user behavior intervention instructions, and the Pareto optimal solution set is selected in combination with user preferences.

[0013] Step 4: Execute the optimized decision plan through the IoT controller and monitor the execution effect in real time;

[0014] Use digital twin technology to build a virtual power system mirror, simulate decision-making effects in extreme scenarios, and dynamically revise risk assessment models and decision-making strategies based on execution feedback data.

[0015] Furthermore, the implementation of the dynamic spatiotemporal weight allocation algorithm in step 1 includes:

[0016] The fusion weight of each sensor is calculated based on the real-time Pearson correlation coefficient between the sensor data and the reference signal, as well as the inverse of the sensor's variance within a preset time window;

[0017] When sensor data is missing, data repair is performed based on the historical co-occurrence probability of adjacent nodes and the neural network prediction model.

[0018] Furthermore, the construction of the spatiotemporal graph convolution-attention hybrid network in step 2 includes:

[0019] Generate a topological adjacency matrix based on the electrical connection relationship of electrical equipment, and use a multi-head attention mechanism to extract the time-dependent characteristics of the equipment status;

[0020] The fuzzy logic algorithm uses an adjustable Gaussian membership function to map risk levels to failure probability.

[0021] Furthermore, the improved non-dominated sorting genetic algorithm in step 3 includes:

[0022] The user-defined preference vector is embedded in the fitness function to screen Pareto solutions that prioritize safety, cost, or power supply continuity.

[0023] In the case of sudden failure, a temporary feasible emergency decision-making plan is generated by relaxing the violation amount of safety constraints or economic constraints.

[0024] Furthermore, the implementation of the closed-loop optimization in step 4 includes:

[0025] A lightweight differential update mechanism is used to synchronize the state differences between the digital twin model and the physical system, reducing communication overhead.

[0026] A causal reasoning model is used to separate the decision execution effect from external interference factors, and to modify the risk assessment model parameters in a targeted manner.

[0027] In a second aspect, the present invention provides a safe electricity use decision-making system, comprising:

[0028] The multi-source data acquisition module is used to collect the operating parameters and environmental parameters of power equipment in real time through sensors deployed in the power network, integrate power grid dispatch data, user historical power consumption behavior data and meteorological data, and build a multi-source heterogeneous data pool;

[0029] Performing edge computing processing on the data, including data cleaning, normalization, and data fusion based on a dynamic spatiotemporal weight allocation algorithm, to generate a dynamic power consumption state matrix;

[0030] The dynamic spatiotemporal weight allocation algorithm dynamically adjusts the fusion weight of each sensor based on the real-time correlation between sensor data and reference signals and the sensor stability score;

[0031] a risk assessment module, configured to input the dynamic power usage state matrix into a deep learning model, identify abnormal patterns in equipment operation in real time, and quantify the risk level based on a fuzzy logic algorithm;

[0032] The deep learning model is a spatiotemporal graph convolution-attention hybrid network, which is used to capture device topology relationships and time series dependency features;

[0033] The risk assessment includes calculating the probability of failure and determining the risk level based on the current over-limit multiple and the temperature rise rate;

[0034] A decision optimization module is used to construct a multi-objective optimization model with safety, energy consumption cost and power supply continuity as optimization objectives based on the risk assessment results;

[0035] An improved non-dominated sorting genetic algorithm is used to globally optimize the control strategy, resource scheduling scheme, and user behavior intervention instructions, and the Pareto optimal solution set is selected in combination with user preferences.

[0036] The closed-loop control module is used to execute the optimized decision-making plan through the IoT controller and monitor the execution effect in real time;

[0037] Use digital twin technology to build a virtual power system mirror, simulate decision-making effects in extreme scenarios, and dynamically revise risk assessment models and decision-making strategies based on execution feedback data.

[0038] Furthermore, the multi-source data acquisition module further includes:

[0039] Edge computing nodes are used to run dynamic spatiotemporal weight allocation algorithms and data completion algorithms in real time;

[0040] In a third aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored in the memory, wherein the processor implements the method when executing the program.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the method described when the program is executed by a processor.

[0042] By implementing the aforementioned technical solutions, a complete technology chain has been established, from precise data collection to closed-loop decision-making optimization. At the data collection end, dynamic weight allocation and topology correlation repair mechanisms ensure high-reliability data even when sensors experience anomalies or communication interruptions, resolving the misjudgment issues often encountered in traditional systems due to single data source failures. At the risk analysis layer, a deep learning model that integrates device physical topology and time series features accurately identifies complex anomalies (such as local overload accompanied by insulation degradation) and outputs interpretable risk levels using fuzzy probability quantification, providing an intuitive basis for decision-making. At the decision-making stage, preference-driven optimization and dynamic constraint adjustment balance safety priorities, energy efficiency costs, and personalized user needs, avoiding resource waste or operational disruptions caused by "one-size-fits-all" strategies. At the feedback iteration layer, lightweight digital twin verification and causal attribution analysis enable rapid verification and targeted correction of policy effectiveness, preventing performance degradation caused by environmental interference or data drift. This complete solution achieves closed-loop improvements in data reliability, risk prediction, decision adaptability, and system self-evolution, significantly outperforming traditional power management systems based on rules or single-objective optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0044] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] Reference Figure 1 The embodiment of the present invention provides a method for generating a safe electricity use decision, which is characterized by comprising the following steps:

[0047] Step 1: Real-time collection and fusion of multi-source heterogeneous data. Sensors deployed in the power network collect real-time operating parameters and environmental parameters of power equipment, integrate grid dispatch data, historical user power consumption behavior data, and meteorological data, and build a multi-source heterogeneous data pool.

[0048] Performing edge computing processing on the data, including data cleaning, normalization, and data fusion based on a dynamic spatiotemporal weight allocation algorithm, to generate a dynamic power consumption state matrix;

[0049] The dynamic spatiotemporal weight allocation algorithm dynamically adjusts the fusion weight of each sensor based on the real-time correlation between sensor data and reference signals and the sensor stability score;

[0050] Step 2: Abnormal pattern recognition and risk assessment: The dynamic power consumption state matrix is ​​input into the deep learning model to identify abnormal patterns in equipment operation in real time and quantify the risk level based on fuzzy logic algorithm;

[0051] The deep learning model is a spatiotemporal graph convolution-attention hybrid network, which is used to capture device topology relationships and time series dependency features;

[0052] The risk assessment includes calculating the probability of failure and determining the risk level based on the current over-limit multiple and the temperature rise rate;

[0053] Step 3: Multi-objective dynamic decision optimization: Based on the risk assessment results, a multi-objective optimization model is constructed with safety, energy consumption cost, and power supply continuity as optimization objectives;

[0054] An improved non-dominated sorting genetic algorithm is used to globally optimize the control strategy, resource scheduling scheme, and user behavior intervention instructions, and the Pareto optimal solution set is selected in combination with user preferences.

[0055] Step 4: Decision execution and closed-loop optimization: Execute the optimized decision plan through the IoT controller and monitor the execution effect in real time;

[0056] Use digital twin technology to build a virtual power system mirror, simulate decision-making effects in extreme scenarios, and dynamically revise risk assessment models and decision-making strategies based on execution feedback data;

[0057] This embodiment describes the above method in detail.

[0058] Step 1: Real-time collection and fusion of multi-source heterogeneous data to achieve full-dimensional perception and high-precision fusion of electrical equipment and environmental data.

[0059] The implementation of the dynamic spatiotemporal weight allocation algorithm includes:

[0060] The fusion weight of each sensor is calculated based on the real-time Pearson correlation coefficient between the sensor data and the reference signal, as well as the inverse of the sensor's variance within a preset time window;

[0061] When sensor data is missing, data repair is performed based on the historical co-occurrence probability of adjacent nodes and the neural network prediction model.

[0062] The specific steps include:

[0063] Step 1.1: Construct an adaptive spatiotemporal weight allocation algorithm to optimize the data fusion quality by dynamically calculating the spatiotemporal weights of each sensor and avoid the error accumulation of the traditional fixed weight algorithm when the device state changes suddenly.

[0064] The sensors used in this embodiment include: a current sensor, a voltage sensor, a temperature and humidity sensor, and a dust concentration sensor.

[0065] In the weight allocation algorithm, the Pearson correlation coefficient (value range [-1,1]) is used to evaluate the linear correlation between sensor data and grid dispatch reference signal within a 5-second time window.

[0066] The weight distribution algorithm is as follows: ;

[0067] Parameter Description:

[0068] :sensor The fusion weight of , the higher the weight, the stronger the data reliability;

[0069] : Calculate sensor by Pearson correlation coefficient Real-time correlation with grid dispatch reference signals;

[0070] : Sensor-based In the time window The stability score is calculated by taking the inverse of the variance within (the smaller the variance, the higher the score);

[0071] : Hyperparameters optimized through grid search, balancing relevance and stability.

[0072] The algorithm runs in real time on the edge computing node, dynamically adjusting the fusion ratio of each sensor data to ensure accurate capture of transient anomalies such as voltage sags and arc faults.

[0073] Through the above algorithm, the data is fused and output to generate a dynamic power consumption status matrix with the dimensions of number of devices × time step × number of features.

[0074] Step 1.2: Data completion based on confidence propagation: For sensor failure or communication interruption scenarios, device topology and historical data are used to repair missing values.

[0075] The completion method is as follows: ;

[0076] Parameter Description: :node With neighbors The historical co-occurrence probability reflects the spatiotemporal correlation; :Prediction model built based on LSTM, input node Historical data (time window is 30 minutes), output missing value estimation; : Historical data of node k is missing.

[0077] When the current sensor fails, the adjacent sensor data (weight ) and historical rules (neural network prediction values) to interpolate missing current values, and combine spatiotemporal correlation confidence with neural network prediction to reduce the interference of missing data on analysis.

[0078] As mentioned above, through the dynamic spatiotemporal weight allocation algorithm, the fusion weight is dynamically adjusted according to the real-time correlation and stability of sensor data and reference signals, which solves the error accumulation problem of traditional fixed weight algorithms when sensors fail or the environment changes suddenly. Combined with the data completion mechanism based on confidence propagation, the device topology relationship and historical data are used to repair missing values, which significantly improves data integrity and fusion accuracy, providing highly reliable input for subsequent analysis.

[0079] Step 2: Deep learning-based abnormal pattern recognition and risk assessment to identify equipment abnormalities in real time and quantify risk probability.

[0080] Among them, a topological adjacency matrix is ​​generated based on the electrical connection relationship of electrical equipment, and the time-dependent characteristics of the equipment status are extracted by combining the multi-head attention mechanism;

[0081] The fuzzy logic algorithm uses an adjustable Gaussian membership function to map risk levels to failure probability.

[0082] Specifically include:

[0083] Step 2.1: Construct a spatiotemporal graph convolution-attention hybrid network (STGC-AN) to integrate the device physical topology and time series features to improve the detection sensitivity of complex abnormal patterns.

[0084] The formula for feature fusion is as follows: ;

[0085] Parameter Description: :Device adjacency matrix, generated according to electrical connection relationship, if the device and Electrical connection, then , otherwise 0; : 8-head attention mechanism to calculate the time-dependent weight of device status; LeakyReLU is the activation function.

[0086] When training the deep learning model, the collected historical fault data set is used for training.

[0087] As described above, the dynamic power consumption status matrix is ​​input into the trained deep learning model, the abnormality score of each device is output, the abnormality of each type of equipment is judged, and the abnormal status of the equipment is quantified.

[0088] Step 2.2: Quantify fuzzy risk probabilistically, converting fuzzy risk levels (such as "high risk") into precise probabilities to support quantitative decision-making.

[0089] The specific quantification method is as follows: ;

[0090] Parameter Description: : adjustable Gaussian membership function, (1.2 times the rated current), ; : Fault severity function, is the weight coefficient, in this embodiment .

[0091] When current exceeds the limit, calculate the probability of a fault occurring in the next 10 minutes ,like , triggering a level one warning.

[0092] As mentioned above, the spatiotemporal graph convolution-attention hybrid network (STGC-AN) is used to synchronously model the topological association and time series dependency characteristics of equipment, enhance the detection sensitivity of complex abnormal patterns (such as arc faults accompanied by temperature rise), and combine with the fuzzy risk probabilistic quantification method to convert fuzzy linguistic variables into precise fault occurrence probabilities, support risk grading and predictive decision-making, and reduce the rates of missed reports and false alarms.

[0093] Step 3: Dynamic decision optimization under multi-objective constraints is used to generate the optimal decision plan for safe, economical and continuous power supply.

[0094] The improved non-dominated sorting genetic algorithm in step 3 includes:

[0095] The user-defined preference vector is embedded in the fitness function to screen Pareto solutions that prioritize safety, cost, or power supply continuity.

[0096] In the case of sudden failure, a temporary feasible emergency decision-making plan is generated by relaxing the violation amount of safety constraints or economic constraints.

[0097] Step 3.1: Screen the Pareto solutions of preference embedding, transforming the user's subjective preferences into constraints for multi-objective optimization to improve the acceptability of decisions.

[0098] Formula and implementation: ;

[0099] Parameter Description: : A user-defined preference vector, corresponding to the weights of safety, energy cost, and power supply continuity (safety: 0.5, cost: 0.3, continuity: 0.2); : Solve the diversity index and avoid local optimality.

[0100] Select the solution with the highest fitness in the Pareto solution set, such as prioritizing reducing the load of high-risk equipment rather than global power rationing.

[0101] Step 3.2: Real-time constraint relaxation algorithm, when the original problem has no solution due to sudden failure, the feasibility is guaranteed by relaxing the constraints.

[0102] The specific implementation formula is: ;

[0103] Parameter Description: : The penalty cost of loose security constraints is higher; : The amount of constraint violation allowed, for example, a temporary overload of 5% is allowed to maintain power supply continuity.

[0104] When the transformer is overloaded and the load cannot be switched immediately, the upper current limit is temporarily relaxed and the backup cooling system is activated.

[0105] As mentioned above, by embedding user preference vectors into an improved non-dominated sorting genetic algorithm (NSGA-II), subjective demands are converted into mathematical constraints, and a Pareto optimal solution set that prioritizes safety, cost, or continuity is screened. Combined with a real-time constraint relaxation strategy, a temporary feasible solution is generated in sudden fault scenarios, balancing safety, economy, and power supply continuity goals, thereby improving the flexibility and operability of decision-making.

[0106] Step 4: Decision execution and feedback-driven closed-loop optimization ensure effective decision execution and continuous model improvement. A lightweight differential update mechanism synchronizes the state differences between the digital twin model and the physical system, reducing communication overhead.

[0107] A causal reasoning model is used to separate the decision execution effect from external interference factors, and to modify the risk assessment model parameters in a targeted manner.

[0108] Specifically include:

[0109] Step 4.1: The incremental synchronization mechanism of the digital twin image reduces the communication and computing overhead of the digital twin system through lightweight differential updates.

[0110] ;

[0111] Parameter Description: : Use SHA-256 algorithm to generate the model state hash value; : An XOR operation locates the difference bits and synchronizes only the changed data. The digital twin model is synchronized every 5 minutes.

[0112] Step 4.2: Feedback correction driven by causal reasoning, distinguishing decision effects from external interference to avoid false corrections.

[0113] The calculation formula for the effect is: ;

[0114] Parameter Description: is the actual observation result; : For decision variables Apply intervention (such as opening a circuit breaker), the specific operation is set by the technicians in charge; : A collection of external factors (such as lightning strikes and human operations); : The probability of external interference e occurring (determined based on historical statistics).

[0115] When the effect deviates from the expected extent and exceeds the preset value, the parameters in the risk assessment are corrected, which is done by those skilled in the art.

[0116] The above-mentioned incremental synchronization mechanism of digital twin mirrors reduces communication overhead, quickly verifies the robustness of decisions in extreme scenarios, separates the decision execution effect from external interference factors through causal reasoning models, and corrects the risk assessment model parameters in a targeted manner to avoid miscorrection and realize dynamic iterative optimization of the system, forming a closed-loop self-evolution capability of "perception-analysis-decision-feedback".

[0117] As described above, the present invention has constructed a complete technology chain from precise data collection to decision-making closed-loop optimization through the technical integration of steps 1 to 4: at the data collection end, through dynamic weight allocation and topology association repair mechanism, it ensures that high-reliability data can still be provided when the sensor is abnormal or the communication is interrupted, solving the misjudgment problem caused by the failure of a single data source in the traditional system; at the risk analysis layer, through the deep learning model that integrates the physical topology and time series characteristics of the equipment, complex anomalies (such as local overload accompanied by insulation degradation) are accurately identified, and the fuzzy probability quantization output can be interpreted to provide an intuitive basis for decision-making; in the decision-making link, through preference-driven optimization and dynamic constraint adjustment, safety red lines, energy efficiency costs and user personalized needs are taken into account to avoid resource waste or operation interruption caused by "one-size-fits-all" strategies; at the feedback iteration layer, through lightweight digital twin verification and causal attribution analysis, rapid verification and targeted correction of strategy effects are achieved to avoid performance degradation of the model due to environmental interference or data drift. The entire solution forms a closed-loop improvement in data reliability, risk prediction capability, decision-making adaptability and system self-evolution capability, and is significantly superior to traditional power management systems based on rules or single-objective optimization.

[0118] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the present invention pertains. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with those in the context of the prior art and, unless specifically defined, will not be interpreted in an idealized or overly formal sense.

[0119] For simplicity of description, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because certain steps can be performed in other orders or simultaneously according to the embodiments of the present invention. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0120] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating safe electricity use decisions, characterized in that: The following steps are involved: Step 1: Sensors deployed in the power network collect real-time operating and environmental parameters of power equipment, integrate grid dispatch data, historical user power consumption data, and meteorological data, and build a multi-source heterogeneous data pool. Performing edge computing processing on the data, including data cleaning, normalization, and data fusion based on a dynamic spatiotemporal weight allocation algorithm, to generate a dynamic power consumption state matrix; The dynamic spatiotemporal weight allocation algorithm dynamically adjusts the fusion weight of each sensor based on the real-time correlation between sensor data and reference signals and the sensor stability score; Step 2: Input the dynamic power consumption state matrix into a deep learning model to identify abnormal patterns in equipment operation in real time and quantify the risk level based on a fuzzy logic algorithm; The deep learning model is a spatiotemporal graph convolution-attention hybrid network, which is used to capture device topology relationships and time series dependency features; The risk assessment includes calculating the probability of failure and determining the risk level based on the current over-limit multiple and the temperature rise rate; Step 3: Based on the risk assessment results, a multi-objective optimization model is constructed with safety, energy consumption cost, and power supply continuity as optimization objectives; An improved non-dominated sorting genetic algorithm is used to globally optimize the control strategy, resource scheduling scheme, and user behavior intervention instructions, and the Pareto optimal solution set is selected in combination with user preferences. Step 4: Execute the optimized decision plan through the IoT controller and monitor the execution effect in real time; Use digital twin technology to build a virtual power system mirror, simulate decision-making effects in extreme scenarios, and dynamically revise risk assessment models and decision-making strategies based on execution feedback data.

2. The method according to claim 1, characterized in that The implementation of the dynamic spatiotemporal weight allocation algorithm in step 1 includes: The fusion weight of each sensor is calculated based on the real-time Pearson correlation coefficient between the sensor data and the reference signal, as well as the inverse of the sensor's variance within a preset time window; When sensor data is missing, data repair is performed based on the historical co-occurrence probability of adjacent nodes and the neural network prediction model.

3. The method according to claim 1, characterized in that The construction of the spatiotemporal graph convolution-attention hybrid network in step 2 includes: Generate a topological adjacency matrix based on the electrical connection relationship of electrical equipment, and use a multi-head attention mechanism to extract the time-dependent characteristics of the equipment status; The fuzzy logic algorithm uses an adjustable Gaussian membership function to map risk levels to failure probability.

4. The method according to claim 1, wherein The improved non-dominated sorting genetic algorithm in step 3 includes: The user-defined preference vector is embedded in the fitness function to screen Pareto solutions that prioritize safety, cost, or power supply continuity. In the case of sudden failure, a temporary feasible emergency decision-making plan is generated by relaxing the violation amount of safety constraints or economic constraints.

5. The method according to claim 1, wherein The implementation of the closed-loop optimization in step 4 includes: A lightweight differential update mechanism is used to synchronize the state differences between the digital twin model and the physical system, reducing communication overhead. A causal reasoning model is used to separate the decision execution effect from external interference factors, and to modify the risk assessment model parameters in a targeted manner.

6. A safe electricity use decision-making system, characterized in that: include: The multi-source data acquisition module is used to collect the operating parameters and environmental parameters of power equipment in real time through sensors deployed in the power network, integrate power grid dispatch data, user historical power consumption behavior data and meteorological data, and build a multi-source heterogeneous data pool; Performing edge computing processing on the data, including data cleaning, normalization, and data fusion based on a dynamic spatiotemporal weight allocation algorithm, to generate a dynamic power consumption state matrix; The dynamic spatiotemporal weight allocation algorithm dynamically adjusts the fusion weight of each sensor based on the real-time correlation between sensor data and reference signals and the sensor stability score; a risk assessment module, configured to input the dynamic power usage state matrix into a deep learning model, identify abnormal patterns in equipment operation in real time, and quantify the risk level based on a fuzzy logic algorithm; The deep learning model is a spatiotemporal graph convolution-attention hybrid network, which is used to capture device topology relationships and time series dependency features; The risk assessment includes calculating the probability of failure and determining the risk level based on the current over-limit multiple and the temperature rise rate; A decision optimization module is used to construct a multi-objective optimization model with safety, energy consumption cost and power supply continuity as optimization objectives based on the risk assessment results; An improved non-dominated sorting genetic algorithm is used to globally optimize the control strategy, resource scheduling scheme, and user behavior intervention instructions, and the Pareto optimal solution set is selected in combination with user preferences. The closed-loop control module is used to execute the optimized decision-making plan through the IoT controller and monitor the execution effect in real time; Use digital twin technology to build a virtual power system mirror, simulate decision-making effects in extreme scenarios, and dynamically revise risk assessment models and decision-making strategies based on execution feedback data.

7. The system according to claim 6, characterized in that The multi-source data acquisition module also includes: Edge computing nodes are used to run dynamic spatiotemporal weight allocation algorithms and data completion algorithms in real time.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images