A multi-module electric meter power management system for adaptive load management

Abstract

The present invention discloses a multi-module electric meter power management system for adaptive load management, comprising: an electric meter module, a load monitoring unit, a data storage unit, and a double-layer optimization control unit. The double-layer optimization control unit includes a particle swarm optimization layer and a fuzzy logic control layer. The particle swarm optimization layer is used to search for the optimal load distribution scheme in a multi-dimensional decision space; the fuzzy logic control layer is used to take the load status and environmental parameters as input variables and generate load migration decisions through fuzzy rule reasoning; the double-layer optimization control unit combines the search ability of the particle swarm optimization layer with the expert knowledge reasoning ability of the fuzzy logic control layer, and dynamically adjusts the weight ratio of the control outputs of the two layers through the adaptive weight function W(t). The present invention provides an adaptive load management system combining a particle swarm optimization system and fuzzy logic control, realizing intelligent allocation and balance of electric loads, and improving the overall efficiency and resource utilization rate of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and in particular to a multi-modular electric meter power management system for adaptive load management. Background Art

[0002] The power management system has been gradually developed since the 1980s, and has undergone significant technological innovation from the initial simple measurement equipment to today's intelligent multi-module system. As an important part of smart grid technology, multi-module meters have gradually become a key node in the modernization of power systems. Early power management mainly adopted manual meter reading and fixed rate billing. By the early 21st century, with the advancement of microprocessor and communication technology, single-function electronic meters were replaced by smart meters with remote data collection and simple load management functions. In recent years, the introduction of modular design concepts has enabled smart meter systems to achieve functional partitioning and on-demand expansion, creating a hardware foundation for adaptive load management.

[0003] The load management of the current multi-modular meter power management system mainly relies on three technical solutions. The static load distribution system is the most traditional method, which distributes the power load to different modules through a preset rule matrix. Make a decision, where P i Represents module priority, L j This method is easy to implement, consumes less resources, and is suitable for scenarios with small load changes, but lacks the ability to cope with dynamic power consumption patterns. The threshold-based dynamic load balancing system sets upper and lower thresholds (T high ,T low ) monitors the module load level and uses Calculate the load migration amount. This method can respond appropriately when the load fluctuates, but its migration decision mechanism is too simplified and fails to fully consider the load characteristics and system efficiency. The predictive load management system uses time series models such as ARMA to predict future load trends, which is mathematically expressed as ,By planning the allocation scheme in advance, it can reduce sudden adjustments.,This system has prediction capabilities, but it relies heavily on historical data,,and has high computational complexity, making it difficult to deal with abnormal power consumption,events.

[0004] Existing systems generally regard each electricity meter module as an independent unit, lacking an effective multi-module cooperation mechanism, resulting in low overall system efficiency. For example, the static allocation system cannot adjust resource allocation in a timely manner when the load fluctuates sharply, causing some modules to be overloaded while others are idle; although the dynamic balancing system can respond to load changes, its simple migration decision-making mechanism is difficult to adapt to complex power usage environments, and the threshold setting and balance factor selection rely too much on experience, and it is difficult to balance system stability and response speed in terms of adjustment frequency. Secondly, existing systems are insufficient in adapting to user behavior patterns and environmental factors and cannot dynamically adjust management strategies according to external factors such as temperature changes and seasonal demands. Moreover, the optimization goal of the system is single, and most only focus on load balancing, ignoring the collaborative optimization of multi-dimensional goals such as power quality and equipment service life. In addition, although the predictive system has a certain degree of practicality, its computational complexity is high and its real-time performance is insufficient. The prediction accuracy is affected by various factors and it cannot effectively cope with sudden power consumption peaks. Summary of the Invention

[0005] The purpose of the present invention is to overcome the technical problems existing in the existing multi-module electricity meter power management system, and provide an adaptive load management system combining a particle swarm optimization system and fuzzy logic control to achieve intelligent allocation and balance of power loads, and improve the overall system efficiency and resource utilization rate. The main design idea of the present invention is to organically combine the search ability of the particle swarm optimization system with the expert knowledge reasoning ability of the fuzzy logic control, and realize the dynamic balance allocation of power loads through a two-layer optimization control structure.

[0006] To achieve the above purpose, the present invention provides a multi-module electricity meter power management system for adaptive load management, which includes: an electricity meter module, a load monitoring unit, a data storage unit, and a two-layer optimization control unit, and the two-layer optimization control unit includes a particle swarm optimization layer and a fuzzy logic control layer.

[0007] The load monitoring unit is mainly responsible for real-time collection of the load level Li of each electricity meter module currenti , where i represents the number of the electricity meter module. This unit real-time monitors the current, voltage and power parameters of each module through a sensor network, calculates the current load level, and transmits the collected data to the central processing unit of the system for subsequent analysis and decision-making.

[0008] The data storage unit is used to store historical load data, user power consumption patterns and environmental parameters. This unit establishes a structured database to record various types of data during the operation of the system, including: historical load change curves of each module, user power consumption behavior characteristics, environmental temperature change records, etc., providing data support for load prediction and decision-making optimization.

[0009] The particle swarm optimization layer is an important part of the double-layer optimization control unit, which is used to search for the optimal load distribution scheme in the multi-dimensional decision space. This layer adopts an improved particle swarm optimization system, and searches for the optimal solution by iteratively updating the particle position vector X and the velocity vector V, where X represents the target load distribution ratio of each module, and V represents the load distribution adjustment rate. The particle swarm optimization system can search for near-optimal solutions in the complex non-linear decision space by simulating swarm intelligence behavior.

[0010] The fuzzy logic control layer is another core component of the double-layer optimization control unit, which is used to take the load status and environmental parameters as input variables and generate load migration decisions through fuzzy rule reasoning. This layer includes a migration timing judgment function F timing and a migration amount calculation function F amount . Through three steps of fuzzification, rule reasoning and defuzzification, expert experience is transformed into specific control decisions. Fuzzy logic control can handle the uncertainty and fuzziness in the system and is suitable for decision-making in complex power load environments.

[0011] The double-layer optimization control unit combines the search ability of the particle swarm optimization layer with the expert knowledge reasoning ability of the fuzzy logic control layer, and dynamically adjusts the weight ratio of the control outputs of the two layers through the adaptive weight function W(t). This combination overcomes the adverse factors that the particle swarm system is prone to fall into local optimum and the fuzzy logic control lacks optimization ability, and realizes the dynamic balanced distribution of the load.

[0012] The particle swarm optimization layer of the present invention adopts an improved particle swarm optimization system, including an adaptive inertia weight factor, particle position and velocity update formulas, and a multi-objective fitness function. The adaptive inertia weight factor w(t) is calculated by the function , where t represents the current iteration number, T max represents the maximum iteration number, w max and w min are the upper and lower limits of the inertia weight respectively, and α is a non-linear adjustment factor. The particle position update formula is , and the particle velocity update formula is , where represents the historical best position of particle i, G best represents the best position found by the group, c1 and c2 are acceleration constants, and r1 and r2 are random numbers in the interval [0,1]. The fitness function f(X) simultaneously considers the load balance degree, power quality and system efficiency, and is expressed as , where β i is the weight coefficient of each sub-objective.

[0013] The fuzzy logic control layer of the present invention includes an input fuzzification unit, a fuzzy rule base, a fuzzy inference mechanism, and an output defuzzification unit. The input fuzzification unit converts the current load level of the electricity meter module , the load change rate , the system load imbalance degree D imbalance and the environmental temperature T into fuzzy sets. The fuzzy rule base contains a set of IF-THEN rules formed by expert knowledge, which is used to handle the complexity and uncertainty of load migration decisions. The fuzzy inference mechanism uses the Mamdani inference method to execute fuzzy rules. The output defuzzification unit converts the fuzzy inference result into a definite load migration timing judgment value F timing and a migration amount calculation value F amount .

[0014] The adaptive weight function W(t) in the present invention dynamically adjusts the output weights of the particle swarm optimization layer and the fuzzy logic control layer according to the system operation state. When the system is in a stable state, W fuzzy (t)>W PSO (t), and the output of the fuzzy logic control layer is preferentially used to reduce the computational overhead; when there are large fluctuations in the system load or the imbalance degree exceeds the preset threshold, W PSO (t)>W fuzzy (t), and the output of the particle swarm optimization layer is preferentially used to obtain the optimal solution; at the same time, W fuzzy (t)+W PSO (t)=1, ensuring weight normalization.

[0015] The present invention further includes a load prediction unit, which predicts the load trend in the future period based on historical load data and calculates the future load using a recursive neural network model , where k is the prediction step, and the prediction result is used as the input of the particle swarm optimization layer and the fuzzy logic control layer to improve the prediction of the decision-making.

[0016] The load balance degree evaluation sub-function f balance (X) in the fitness function of the present invention is calculated as the ratio of the standard deviation of the load levels of each module to the average load: , where σ(L current ) represents the standard deviation of the current loads of each module, and L average represents the average load level. The smaller the value of this sub-function, the more balanced the load distribution.

[0017] The fuzzy rule base of the present invention contains professional rules for different load scenarios, specifically including: a rule set for peak load scenarios, used for rapid response and load dispersion; a rule set for valley load scenarios, used for device efficiency and energy conservation; a rule set for rapidly fluctuating load scenarios, used for system stability and buffering capacity; a rule set for seasonal variations, dynamically adjusting the load distribution strategy according to the environmental temperature T and historical data of the same period.

[0018] The present invention sets up a multi-level load migration priority mechanism. By comprehensively considering the current load level of the electricity meter module , the historical load fluctuation amplitude Var(L i ) and the device health status index H i , the source module and target module of load migration are determined. Load is preferentially migrated from a high-load and healthy module to a low-load module. The load migration priority calculation formula is: , where γ1, γ2, and γ3 are weight coefficients.

[0019] The present invention realizes a collaborative learning mechanism between the particle swarm optimization system and fuzzy logic control, specifically manifested as: the search results of the particle swarm optimization system are used to dynamically adjust the membership function parameters of the fuzzy rules; the expert knowledge of fuzzy control is used to guide the initialization and mutation operations of the particle swarm; through collaborative learning, the system can simultaneously overcome the disadvantages of slow convergence speed of the particle swarm system and difficult design of the fuzzy logic rule base.

[0020] The present invention also includes an emergency response module. When abnormal power grids or sudden load changes are detected, it bypasses the conventional decision-making process and directly adopts a preset emergency load distribution strategy to ensure the robustness and safety of the system in extreme situations. The emergency response trigger conditions are: , where ε is the load imbalance threshold and η is the load change rate threshold.

[0021] In summary, by combining the particle swarm optimization system and fuzzy logic control technology, the present invention constructs a two-layer optimization control structure and realizes the adaptive performance optimization of the multi-module electricity meter power management system, having the following beneficial effects:

[0022] 1. Improve the accuracy and balance of load distribution: Traditional static load distribution systems and threshold-based dynamic load balancing systems lack accuracy when facing complex power load environments. Through the search ability of the particle swarm optimization system, the present invention can find a nearly optimal load distribution plan in the multi-dimensional decision space, improving the load balance degree. Compared with traditional methods, the load balance index f balance (X) of this system decreases, effectively reducing the risk of module overload.

[0023] 2. Enhanced the system's adaptability to dynamic changes: The existing load management system has insufficient response ability to changes in the power consumption environment. The present invention combines fuzzy logic control technology, transforms expert experience into adaptive decision rules, and can dynamically adjust the load distribution strategy according to real-time load status, power consumption patterns, and environmental parameters. The response time of the system to sudden load changes is shortened, and the adaptability to seasonal load changes is improved.

[0024] 3. Achieved multi-objective collaborative optimization: Traditional systems often focus on a single optimization goal. The present invention constructs a multi-dimensional evaluation system, takes into account load balance, power quality, and system efficiency at the same time, and realizes the collaborative balance of optimization goals through a multi-objective fitness function. While ensuring load balance, the system efficiency is improved, and the power quality index is improved.

[0025] 4. Improved the balance between computational efficiency and optimization accuracy: Although the particle swarm optimization system has a powerful search ability, its computational complexity is high; although fuzzy logic control has high computational efficiency, its optimization accuracy is limited. The present invention dynamically adjusts the output weights of the two systems through the adaptive weight function W(t). When the system is stable, it preferentially uses fuzzy logic control with high computational efficiency, and switches to the particle swarm system with high optimization accuracy when the load fluctuates violently, realizing the dynamic balance between computational efficiency and optimization accuracy. The average computational load of the system is reduced, while maintaining a high optimization accuracy.

[0026] 5. Enhanced the reliability of the system: By setting up an emergency response module and a multi-level load migration priority mechanism, this system can effectively respond to power grid anomalies and sudden load changes, ensuring the stable operation of the system under extreme conditions. The failure rate of the system when facing sudden load peaks is reduced, and the system instability caused by power grid fluctuations is reduced.

[0027] 6. Achieved the self-learning and evolution of the system: The collaborative learning mechanism of the present invention enables the particle swarm optimization system and fuzzy logic control to promote each other and co-evolve. The search results of the particle swarm optimization system are used to dynamically adjust the membership function parameters of the fuzzy rules, improving the accuracy of fuzzy control; the expert knowledge of fuzzy control is used to guide the initialization and mutation operations of the particle swarm, accelerating the convergence speed of the particle swarm system. This improves the system optimization performance and enhances the adaptability to new load patterns.

[0028] 7. Improved the overall system efficiency and resource utilization rate: Through the adaptive load management solution of the present invention, the resource utilization rate of the meter module is improved, the peak-valley difference of the power grid load is reduced, the operating energy consumption of the system is reduced, the service life of the equipment is extended, and the comprehensive economic benefits are significantly improved.

[0029] In summary, by combining the particle swarm optimization system and the fuzzy logic control technology, the present technical solution constructs a load management system with adaptive ability, collaborative learning ability and multi-objective optimization ability, effectively overcoming the limitations of traditional load management systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The present invention will be further described in detail below in conjunction with specific embodiments, but the embodiments of the present invention are not limited thereto.

[0032] Referring to Figure 1 As shown, the multi-module electric meter power management system for adaptive load management of the present invention mainly includes: an electric meter module, a load monitoring unit, a data storage unit, a double-layer optimization control unit, a load prediction unit and an emergency response module. The system connects each functional unit through a bus and is coordinated by a central processing unit.

[0033] The load monitoring unit adopts a distributed sensing network structure, and real-time collects the electrical parameters of each electric meter module through high-precision current transformers and voltage sensors, including current (I, unit: ampere), voltage (U, unit: volt), power factor (cosφ), etc., and calculates the current load level (unit: kilowatt). After the collected data is filtered and calibrated, it is transmitted to the central processing unit in a standard format. The sampling frequency of the load monitoring unit can be dynamically adjusted according to system requirements, usually set to once per minute in the stable state and can be increased to multiple times per second during load fluctuations.

[0034] The data storage unit adopts a hierarchical architecture, including a real-time data buffer and a historical database. The real-time data buffer uses high-speed memory to store the load data of the most recent 24 hours; the historical database uses a relational database to store long-term operation data, including the historical load curves of each module (unit: kilowatt, t is time), the user power consumption mode feature vector P user (including parameters such as power consumption period, power consumption, power consumption stability, etc.), and environmental parameters E (including temperature T, humidity H, air pressure P, etc.). The data storage unit is also responsible for data compression, backup and recovery functions to ensure system data security.

[0035] The double-layer optimization control unit is the core technical solution of the present invention, including a particle swarm optimization layer, a fuzzy logic control layer and an adaptive weight adjustment mechanism.

[0036] The particle swarm optimization layer uses an improved particle swarm optimization system to search for the optimal solution in the load distribution decision space. The specific implementation steps of this layer are as follows:

[0037] 1. Initialize the particle swarm: Initialize m particles in the N-dimensional decision space, where N is equal to the number of electricity meter modules, and m is usually set to 10×N. Each particle represents a possible load distribution scheme, and its position vector X i represents the target load distribution ratio of each module, and the velocity vector V i represents the load distribution adjustment rate. The initial position and velocity are randomly generated using a uniform distribution, and the position vector satisfies the normalization constraint: .

[0038] 2. Evaluate the particle fitness: Calculate the fitness value f(X i ) for each particle. This fitness function includes three aspects: load balance, power quality, and system efficiency:

[0039] Load balance sub-function , where σ(L current ) represents the standard deviation of the current load of each module (unit: kilowatt), and L average represents the average load level (unit: kilowatt). The smaller this value, the more balanced the load distribution.

[0040] Power quality sub-function f quality (X) considers voltage stability, harmonic content, and power factor, and its calculation formula is , where VFI is the voltage fluctuation index (dimensionless), THD is the total harmonic distortion rate (%), PF is the power factor, and w1, w2, w3 are weight coefficients.

[0041] System efficiency sub-function f efficiency (X) considers energy utilization rate and operating cost, and its calculation formula is , where E loss is the energy loss (unit: kilowatt-hour), E total is the total electricity consumption (unit: kilowatt-hour), C operation is the operating cost (unit: yuan), and C max is the preset maximum cost (unit: yuan).

[0042] Total fitness function:

[0043] , where β1, β2, β3 are weight coefficients, and β1 + β2 + β3 = 1. The initial values are set to β1 = 0.5, β2 = 0.3, β3 = 0.2, and can be dynamically adjusted according to actual application requirements.

[0044] 3. Update the individual optimal solution and the optimal solution: Compare the current fitness of each particle with its historical best fitness, and update the individual optimal position P best ; Compare the fitness of all particles and update the global optimal position Gbest 。

[0045] 4. Update particle velocity and position: Use improved velocity and position update formulas, introducing an adaptive inertia weight and a convergence factor:

[0046] Adaptive inertia weight , where t is the current iteration number, T max is the maximum number of iterations, usually set to 100 - 200, w max and w min are the upper and lower limits of the inertia weight, usually set to 0.9 and 0.4 respectively, and α is a non-linear adjustment factor, usually set to 1.2.

[0047] Particle velocity update formula:

[0048] , where c1 and c2 are acceleration constants, usually set to 2.0, and r1 and r2 are random numbers in the range [0, 1].

[0049] Particle position update formula: 。

[0050] Position constraint handling: The updated position vector may not satisfy the normalization constraint, and normalization processing is required: 。

[0051] 5. Convergence judgment: Check whether the iteration reaches the maximum number of times T max , or whether the improvement amplitude of the optimal solution in consecutive iterations is lower than the preset threshold ε convergence (usually set to 0.001). If either condition is met, stop the iteration; otherwise, return to step 2 to continue the iteration.

[0052] The disadvantages of the particle swarm optimization layer mainly include: high computational complexity, especially the significant iterative calculation overhead in systems with a large number of modules; being prone to falling into local optimal solutions and difficult to find the true optimal; being sensitive to parameter settings, and different parameter combinations may lead to significant differences in system performance; slow convergence speed, requiring more iterations to achieve satisfactory results.

[0053] The fuzzy logic control layer converts expert experience into specific load migration decisions through fuzzy rule reasoning. The specific implementation steps of this layer are as follows:

[0054] 1. Fuzzification of input variables: Convert the current load level of the electricity meter module , load change rate , system load imbalance degree D imbalance and environmental temperature T into fuzzy sets. Each input variable is defined with 5 - 7 fuzzy sets, specifically as follows:

[0055] Load level L currenti : Very low (VL), low (L), medium (M), high (H), very high (VH).

[0056] Load change rate : Fast decline (FD), slow decline (SD), stable (S), slow increase (SI), fast increase (FI).

[0057] Load imbalance degree D imbalance : Balanced (B), slightly unbalanced (SU), moderately unbalanced (MU), severely unbalanced (HU), extremely unbalanced (VU).

[0058] Ambient temperature T: Very cold (VC), cold (C), moderate (M), hot (H), very hot (VH).

[0059] Each fuzzy set is described using a triangular or trapezoidal membership function μ(x). The triangular membership function is defined as:

[0060]

[0061] where a, b, c are membership function parameters, satisfying a ≤ b ≤ c.

[0062] 2. Construct a fuzzy rule base: Based on the expert knowledge of the power system, construct a fuzzy rule base containing multiple sets of rule sets. The rules adopt the IF-THEN structure.

[0063] Rules for peak load scenarios:

[0064]

[0065] Rules for valley load scenarios:

[0066]

[0067] Rules for fast fluctuating load scenarios:

[0068]

[0069] Rules for seasonal variations:

[0070]

[0071] The rule settings in the rule base can cover all possible load scenarios and environmental conditions.

[0072] 3. Fuzzy inference: The Mamdani inference method is used to execute the fuzzy rules. For each rule, first calculate the membership degree of the antecedent part, and then determine the triggering strength of the consequent part. The specific steps are as follows:

[0073] Calculate the membership degree of the antecedent part of the calculation rule: For conditions connected by AND, take the minimum value; for conditions connected by OR, take the maximum value.

[0074] Determine the triggering strength of the consequent part of the rule: Use the truncation method, and take the membership degree of the antecedent part as the maximum height of the fuzzy set of the consequent part.

[0075] Aggregate the outputs of all rules: Use the maximum method to combine the output fuzzy sets of all rules.

[0076] 4. Output defuzzification: Convert the fuzzy inference result into a definite judgment value F for the load migration timing timing and a calculated value F for the migration amount amount . Use the centroid method for defuzzification: ; where μ(x) is the membership function of the aggregated output fuzzy set, x is the value of the output variable, and F is the output value after defuzzification.

[0077] F timing The output is a value within the interval [0, 1], representing the urgency of load migration. 0 means no migration is required, and 1 means immediate migration.

[0078] F amount The output is a value within the interval [0, 1], representing the relative amount of load migration. The actual migration amount ΔL = F amount ×L max _ transfer where L max _ transfer is the maximum allowable migration load of the system (unit: kilowatt).

[0079] The disadvantages of the fuzzy logic control layer mainly include: The design of the rule base depends on expert experience and it is difficult to cover all possible scenarios; The setting of the parameters of the membership function lacks theoretical guidance and mostly relies on the trial-and-error method to determine; It lacks the ability to optimize and it is difficult to ensure the optimality of the decision-making; The number of rules is proportional to the system complexity, and the rule base of a complex system may be too large to be maintained.

[0080] The dynamic adjustment mechanism of the adaptive weight function W(t) dynamically adjusts the output weights of the particle swarm optimization layer and the fuzzy logic control layer according to the system operation state to achieve the collaborative work of the two systems. The specific implementation is as follows:

[0081] 1. System state evaluation: Calculate the system state index S(t) based on the load monitoring data, including:

[0082] Load fluctuation index , representing the ratio of the maximum load change to the average load.

[0083] Load imbalance index , representing the ratio of the load standard deviation to the average load.

[0084] System state index , where λ1 and λ2 are weight coefficients, satisfying λ1 + λ2 = 1.

[0085] 2. Weight function calculation: Calculate the weight function value based on the system state index S(t):

[0086] When S(t) ≤ S threshold_low , the system is in a stable state,

[0087] Set W fuzzy (t) = W fuzzy_max , W PSO (t) = 1 - W fuzzy (t).

[0088] When S(t) ≥ S threshold_high , the system is in a fluctuating state,

[0089] Set W PSO (t) = W PSO_max , W fuzzy (t) = 1 - W PSO (t).

[0090] When S threshold_low < S(t) < S threshold_high , linearly interpolate to calculate the weight:

[0091] W fuzzy (t) = W fuzzy_max - (W fuzzy_max - W fuzzy_min ) × (S(t) - S threshold_low ) / (S threshold_high - S threshold_low );

[0092] W PSO (t) = 1 - W fuzzy (t).

[0093] Where S threshold_low and S threshold_high are respectively the lower and upper limits of the system state threshold, W fuzzy_max and W fuzzy_min are respectively the upper and lower limits of the weights in the fuzzy logic control layer, W PSO_max and W PSOmin are respectively the upper and lower limits of the weights in the particle swarm optimization layer. Typical parameter settings are: S threshold_low = 0.1, S threshold_high = 0.3, W fuzzy_max = 0.8, W fuzzy_min = 0.2, W PSO_max = 0.8, WPSOmin = 0.2.

[0094] 3. Control output fusion: The outputs of the particle swarm optimization layer and the fuzzy logic control layer are fused into the final control decision according to the weight function:

[0095] Load migration timing: F timing_final = W fuzzy (t) × F timingfuzzy + W PSO (t) × F timingPSO ;

[0096] Load migration amount: F amount _final = W fuzzy (t) × F amountfuzzy + W PSO (t) × F amountPSO .

[0097] The present invention realizes a collaborative learning mechanism for a particle swarm optimization system and fuzzy logic control, specifically manifested as:

[0098] 1. Particle swarm optimization results guide fuzzy rule adjustment:

[0099] Record the optimal solution X of the particle swarm optimization system under different load scenarios opt and its corresponding system state parameters.

[0100] Analyze the difference ΔF = F between the optimal solution and the output of the current fuzzy rule PSO - F fuzzy . - Adjust the membership function parameters of the fuzzy rule according to the difference: a' = a + η·ΔF, b' = b + η·ΔF, c' = c + η·ΔF, where η is the learning rate, usually set to 0.05 - 0.1.

[0101] Regularly evaluate the performance of the adjusted rules, retain the adjustments with performance improvement, and discard the adjustments with performance decline.

[0102] 2. Fuzzy control guides particle swarm system optimization:

[0103] Use the expert knowledge of fuzzy control to guide the initialization of the particle swarm: Generate initial particles close to the potential optimal solution according to the current system state and the inference results of the fuzzy rules, and accelerate the system convergence.

[0104] Use fuzzy rules to evaluate the diversity of the particle swarm. When the diversity is insufficient, trigger the mutation operation: Randomly perturb the positions of some particles to avoid falling into local optima.

[0105] Construct a speed constraint mechanism based on fuzzy rules: Adjust the speed upper limit V according to the system state max, reduce the speed limit in the critical state to improve the search accuracy, and increase the speed limit in the general state to accelerate convergence.

[0106] 3. Collaborative learning performance evaluation and adjustment:

[0107] Regularly evaluate the effect of collaborative learning, calculate the system performance index P(t), including load balance degree, response time, energy efficiency, etc.

[0108] Dynamically adjust the learning parameters according to the performance change trend ΔP = P(t) - P(t - T), including the learning rate η, mutation probability p mutation , speed constraint coefficient, etc.

[0109] Build a knowledge base to record successful learning cases as a reference basis for self-learning.

[0110] Through the collaborative learning mechanism, the system can simultaneously overcome the disadvantages of slow convergence speed of the particle swarm system and difficult design of the fuzzy logic rule base, and achieve continuous improvement of system performance.

[0111] The emergency response module is a key component to ensure the safe and stable operation of the system in extreme situations, and the specific implementation is as follows:

[0112] 1. Abnormal state detection: Continuously monitor key system parameters, including:

[0113] Load imbalance degree:

[0114] Load change rate:

[0115] Voltage deviation:

[0116] Temperature anomaly:

[0117] 2. Emergency trigger condition: Trigger an emergency response when any parameter exceeds the preset threshold:

[0118] > ε, where ε is the load imbalance threshold, usually set to 0.3 - 0.5. > η, where η is the load change rate threshold, usually set to 0.2 - 0.3 kW / min.

[0119] > δ, where δ is the voltage deviation threshold, usually set to 0.1.

[0120] > τ, where τ is the temperature anomaly threshold, usually set to 15 °C.

[0121] 3. Emergency Response Strategy: After triggering the emergency response, bypass the regular decision-making process and directly execute the preset strategy:

[0122] Load Balancing Strategy: Rapidly reallocate the load to make the loads of each module tend to the average value L average .

[0123] Load Limitation Strategy: Implement load limitation on overloaded modules to ensure that the load does not exceed the safety threshold L safe_i .

[0124] Standby Module Startup Strategy: Start the standby electricity meter module in extreme cases to share the system load.

[0125] Fault Isolation Strategy: Identify and isolate faulty modules, and reorganize the remaining modules to form a new operating network.

[0126] 4. Recovery Mechanism: After the abnormal situation is resolved, the system smoothly transitions back to the regular control mode:

[0127] Calculate the transition period length T transition , usually 5 - 10 minutes.

[0128] During the transition period, linearly reduce the emergency response weight and increase the regular control weight.

[0129] After the transition period ends, fully restore the regular control mode.

[0130] The load prediction unit uses a recurrent neural network model to predict future loads, improving the prediction of system decisions. The specific implementation is as follows:

[0131] 1. Data Preprocessing: Time Window Selection: Select an appropriate time window length according to the prediction target. Usually, a 10 - 30 - minute window is selected within 1 hour, a 4 - 6 - hour window within 1 day, and a 3 - 7 - day window within 1 week.

[0132] Feature Extraction: In addition to the load historical data, time features, temperature data, user behavior features, etc. are also extracted as inputs. Data Standardization: Standardize all input features and map them to the interval [0, 1] or [-1, 1].

[0133] 2. Model Structure:

[0134] Adopt the long short - term memory network (LSTM) as the core prediction model, including an input layer, an LSTM hidden layer (1 - 2 layers), and an output layer.

[0135] The number of nodes in the input layer is equal to the number of input features, usually 10 - 20 nodes. Each LSTM hidden layer contains 30 - 50 neurons and uses the tanh activation function. The number of nodes in the output layer is equal to the prediction step k, usually 6 - 12 nodes (predicting the next 1 - 2 hours, with a prediction point every 10 minutes).

[0136] 3. Model training: The sliding window method is adopted to generate training samples, and the window step size is 10 minutes.

[0137] The batch size is set to 32 - 64, and the number of training epochs is 100 - 200.

[0138] The Adam optimizer is used, the initial learning rate is set to 0.001, and a learning rate decay strategy is adopted.

[0139] The mean squared error (MSE) is used as the loss function, and the mean absolute percentage error (MAPE) is monitored simultaneously to evaluate the prediction accuracy.

[0140] 4. Prediction result processing:

[0141] Calculate the prediction confidence interval: ±1.96σ prediction , where σ prediction is the prediction standard deviation.

[0142] Anomaly prediction detection: If the predicted value deviates too much from the historical data of the same period, a warning is triggered and the prediction result is adjusted.

[0143] Smoothing processing of prediction results: Apply a moving average filter to the prediction curve to reduce abnormal fluctuations.

[0144] 5. Application of prediction results: The short-term prediction results are directly used as the input for the particle swarm optimization layer and the fuzzy logic control layer. For medium- and long-term prediction results, those over 1 hour are used for system resource planning and peak-valley load pre-adjustment. Prediction error feedback: The deviation between the actual load and the predicted load is used for continuous optimization of the model.

[0145] The multi-level load migration priority mechanism determines the source module and target module of load migration by comprehensively considering various factors, and the specific implementation is as follows:

[0146] 1. Calculation of migration priority:

[0147] Current load level L currenti : The real-time load level of the module, in kilowatts.

[0148] Historical load fluctuation amplitude Var(L i ): The variance of the load within 24 hours, reflecting the load stability.

[0149] Device health status index H i:The health status index calculated based on device running time, fault records, temperature, etc., with a value range of [0, 1], where 1 represents complete health.

[0150] Migration priority calculation formula: , where γ1, γ2, γ3 are weight coefficients, satisfying γ1 + γ2 + γ3 = 1, and the typical setting is γ1 = 0.6, γ2 = 0.3, γ3 = 0.1.

[0151] 2. Migration source module selection:

[0152] For all modules, sort them in descending order, and select the top N source (usually 20% - 30% of the total number of modules) of the modules as candidate source modules.

[0153] Check whether the load level of the candidate source modules exceeds the migration threshold L threshold_source , usually set as L average ×(1 + δ source ), where δ source is the source module threshold coefficient, which is 0.2.

[0154] Finally, determine the modules that meet the conditions as the actual source module set S source .

[0155] 3. Migration target module selection:

[0156] For all modules, sort them in ascending order, and select the top N target (usually 20% - 30% of the total number of modules) of the modules as candidate target modules. - Check whether the load level of the candidate target modules is lower than the migration threshold L threshold_target , usually set as L average ×(1 - δ target ), where δ target is the target module threshold coefficient, usually 0.2.

[0157] Check whether the remaining capacity of the candidate target modules is sufficient to receive the migration load. The remaining capacity is calculated as C remaini = L maxi - L currenti , where L maxi is the maximum load capacity of module i. - Finally, determine the modules that meet the conditions as the actual target module set S target .

[0158] 4. Migration volume calculation and allocation:

[0159] Calculate the total migration volume of the source module set: L migration_total = ∑(L current_i - Lbalance_i ), where i ∈ S source , L balance_i is the target balanced load of module i, usually set to L average .

[0160] Calculate the total receiving capacity of the target module set: C receive_total = ∑(L balancei - L currenti ), where i ∈ S target . - If L migration_total > C receive_total , allocate the migration amount proportionally; if L migration_total ≤ C receive_total , allocate the migration amount completely.

[0161] For each source module i and target module j, calculate the migration amount L migration_i ,j, and optimize the migration path considering factors such as network topology and distance cost.

[0162] The multi - modular electric meter power management system with adaptive load management proposed by the present invention constructs a two - layer optimization control structure by combining the particle swarm optimization system and fuzzy logic control technology, realizing the intelligent allocation and balance of power loads. The system uses an adaptive weight function to dynamically adjust the output weights of the two systems, reducing the computational overhead while ensuring the optimization accuracy. Through the collaborative learning mechanism, the system can continuously improve the fuzzy rules and optimize the system performance, achieving continuous evolution. The multi - level load migration priority mechanism and emergency response module further enhance the flexibility and reliability of the system.

[0163] Compared with traditional static load distribution systems, threshold - based dynamic load balancing systems, and predictive load management systems, the solution provided by the present invention has significant advantages: more balanced load distribution, stronger adaptability to dynamic changes, ability to achieve multi - objective collaborative optimization, higher system efficiency and resource utilization rate. The present invention is applicable to various application scenarios such as intelligent buildings, industrial parks, intelligent communities, new energy micro - grids, and electric vehicle charging stations, and has broad application prospects.

[0164] To better understand the present invention, this application provides an example calculation process of a multi - modular electric meter power management system with adaptive load management:

[0165] 1. Scenario setting

[0166] This case simulates a multi - modular electric meter load management system in a large commercial building. The building has 6 electric meter modules, each responsible for different functional areas:

[0167] Module 1: Office area

[0168] Module 2: Commercial Retail Area

[0169] Module 3: Dining Area

[0170] Module 4: Public Facilities, Elevators, Lighting, etc.

[0171] Module 5: Air Conditioning System

[0172] Module 6: Backup Power Module

[0173] Assumption: At 14:00 on a summer weekday, the outdoor temperature reaches 35°C, the air conditioning load suddenly increases, and at the same time, the electricity demand in the commercial retail area increases due to promotional activities. The system needs to perform load management to avoid overloading of individual modules.

[0174] 2. Data Input

[0175] 2.1 Current Load Data of Each Module

[0176] Module number <![CDATA[Current load L current， (kW)]]> Load change rate (kW / min) <![CDATA[Maximum load capacity L max (kW)]]> <![CDATA[Historical load fluctuation range Var(L i )]]> <![CDATA[Device Health Status Index H i > 1 120 +0.5 200 15 0.95 2 180 +2.5 220 25 0.90 3 90 -0.2 150 20 0.85 4 110 +0.3 180 10 0.98 5 210 +3.0 230 35 0.92 6 40 +0.1 200 5 1.00

[0177] 2.2 System Parameters

[0178] Total System Load: L total = 750 kW

[0179] Average Load: L average = 125 kW

[0180] System Load Standard Deviation: σ(L current ) = 58.14 kW

[0181] Load Imbalance Degree: D imbalance = σ(L current ) / L average = 0.465

[0182] Ambient Temperature: T = 35°C

[0183] Predicted 1-hour Load Trend: Continuously Increasing, L predicted_t+60 = 820 kW

[0184] 2.3 System Parameter Settings

[0185] 1. Dynamic Load Balancing System Parameters Based on Thresholds:

[0186] High Load Threshold: T high = L average × 1.2 = 150 kW

[0187] Low Load Threshold: T low = L average × 0.8 = 100 kW

[0188] Balancing Factor: α = 0.3

[0189] 2. Particle swarm optimization system parameters:

[0190] Number of particles: m = 30

[0191] Maximum number of iterations: T max = 100

[0192] Inertia weight range: w max = 0.9, w min = 0.4

[0193] Nonlinear adjustment factor: α = 1.2

[0194] Acceleration constants: c1 = c2 = 2.0

[0195] Fitness function weights: β1 = 0.6 (load balance degree), β2 = 0.3 (power quality), β3 = 0.1 (system efficiency)

[0196] 3. Fuzzy logic control parameters:

[0197] Input variable membership functions:

[0198] Load level L currenti : Very low (VL), low (L), medium (M), high (H), very high (VH)

[0199] Load change rate ΔL i / Δt: Fast decline (FD), slow decline (SD), stable (S), slow rise (SI), fast rise (FI)

[0200] Load imbalance degree D imbalance : Balanced (B), slightly unbalanced (SU), moderately unbalanced (MU), severely unbalanced (HU), extremely unbalanced (VU)

[0201] Ambient temperature T: Very low (VC), low (C), moderate (M), high (H), very high (VH).

[0202] 4. Adaptive weight function parameters:

[0203] System state threshold: S threshold_low = 0.1, S threshold_high = 0.3

[0204] Weight range: W fuzzy_max = 0.8, W fuzzy_min = 0.2, W PSO_max = 0.8, W PSOmin = 0.2

[0205] State evaluation weight: λ1 = 0.4 (load fluctuation index), λ2 = 0.6 (load imbalance index)

[0206] 5. Multi-level load migration priority parameter:

[0207] Weight coefficient: γ1 = 0.6, γ2 = 0.3, γ3 = 0.1

[0208] System state evaluation and adaptive weight function calculation

[0209] 1. Calculate the load fluctuation index:

[0210] LFI(t)=max|ΔL i / Δt| / L average =3.0 / 125=0.024

[0211] 2. Calculate the load imbalance index:

[0212] LUI(t)=σ(L current ) / L average =58.14 / 125=0.465

[0213] 3. Calculate the system state index:

[0214] S(t)=λ1·LFI(t)+λ2·LUI(t)=0.4×0.024+0.6×0.465=0.0096+0.279=0.289

[0215] 4. Since S threshold_low <S(t)<S threshold_high (0.1<0.289<0.3),

[0216] Calculate the adaptive weight:

[0217] W fuzzy (t)=W fuzzy_max -(W fuzzy_max -W fuzzy_min )×(S(t)-S threshold_low ) / (S threshold_high -S threshold_high -S threshold_low )=0.8-(0.8-0.2)×(0.289-0.1) / (0.3-0.1)=0.8-0.6×0.945=0.8-0.567=0.233;

[0218] W PSO (t)=1-W fuzzy (t)=1-0.233=0.767。

[0219] 5. Weight analysis: Since the system load imbalance degree is relatively high, approaching S threshold_high , the system preferentially adopts the output of the particle swarm optimization layer (W PSO (t)=0.767) to obtain a solution closer to the optimal one.

[0220] The fuzzy logic control layer calculates

[0221] 1. Fuzzification of input variables:

[0222] Taking module 5 as an example, its current load L current =210kW, membership degree calculation: μVL(210)=0, μL(210)=0, μM(210)=0, μH(210)=0.2, μVH(210)=0.8

[0223] The load change rate ΔL5 / Δt = +3.0kW / min,

[0224] Membership degree calculation:

[0225] μFD(3.0)=0, μSD(3.0)=0, μS(3.0)=0, μSI(3.0)=0.1, μFI(3.0)=0.9 - System load imbalance degree D imbalance =0.465,

[0226] Membership degree calculation:

[0227] μB(0.465)=0, μSU(0.465)=0, μMU(0.465)=0.3, μHU(0.465)=0.7, μVU(0.465)=0 - Ambient temperature T = 35℃.

[0228] Membership degree calculation:

[0229] μVC(35)=0, μC(35)=0, μM(35)=0, μH(35)=0.3, μVH(35)=0.7

[0230] 2. Activation of fuzzy rules:

[0231] Rule 1

[0232] IF (L currenti is VH) AND (ΔL i / Δt is FI) AND (D imbalance is HU) AND (T is VH) THEN (F timing is Immediate) AND (F amount is Large);

[0233] Rule strength = min(0.8, 0.9, 0.7, 0.7) = 0.7

[0234] Rule 2:

[0235] IF (L currenti isH) AND (ΔL i / Δt is SI) AND (D imbalance isMU) THEN (F timing isSoon) AND (F amount isMedium);

[0236] Rule strength = min(0.2, 0.1, 0.3) = 0.1;

[0237] Rule 3:

[0238] IF (TisVH) AND (L currenti isVH) THEN (F timing isImmediate) AND (F amount isMedium);

[0239] Rule strength = min(0.7, 0.8) = 0.7

[0240] 3. Defuzzification calculation - simplified to weighted average:

[0241] F timingfuzzy = (0.7×1.0 + 0.1×0.8 + 0.7×1.0) / (0.7 + 0.1 + 0.7) = 1.47 / 1.5 = 0.98

[0242] F amountfuzzy = (0.7×0.8 + 0.1×0.5 + 0.7×0.5) / (0.7 + 0.1 + 0.7) = 0.91 / 1.5 = 0.61

[0243] 4. Fuzzy control output result:

[0244] Emergency level of load migration for Module 5: F timingfuzzy = 0.98 (Very urgent)

[0245] Relative amount of load migration for Module 5: F amountfuzzy = 0.61

[0246] Actual migration amount:

[0247] ΔL fuzzy = F amountfuzzy ×(L current5 - L average ) = 0.61×(210 - 125) = 51.85 kW

[0248] Particle Swarm Optimization Layer Calculation

[0249] 1. Particle Initialization: Generate 30 random particles, each particle representing a load distribution scheme. Taking a representative particle as an example: X i = [0.16, 0.17, 0.15, 0.16, 0.18, 0.18] (normalized load distribution ratio)

[0250] 2. Particle Evaluation (taking the initial representative particle as an example):

[0251] The load distribution corresponding to this particle:

[0252] L PSO = X i ×L total = [120, 127.5, 112.5, 120, 135, 135]

[0253] Load Balancing Degree: f balance (X i ) = σ(L PSO ) / L average = 8.37 / 125 = 0.067

[0254] Assume Power Quality Evaluation: f quality (X i ) = 0.15 - Assume System Efficiency Evaluation: f efficiency (X i ) = 0.08

[0255] Comprehensive Fitness:

[0256] f(X i ) = 0.6×0.067 + 0.3×0.15 + 0.1×0.08 = 0.0402 + 0.045 + 0.008 = 0.0932

[0257] 3. Iterative Optimization:

[0258] The optimal particle at the 100th iteration:

[0259] X best = [0.156, 0.173, 0.165, 0.158, 0.184, 0.164]

[0260] The corresponding load distribution:

[0261] L PSO = X best ×L total = [117, 129.75, 123.75, 118.5, 138, 123]

[0262] Final fitness: f(X best ) = 0.047 (significantly reduced compared to the initial value of 0.0932)

[0263] 4. Output results of particle swarm optimization:

[0264] Module 1: 117 kW (decrease of 3 kW)

[0265] Module 2: 129.75 kW (decrease of 50.25 kW)

[0266] Module 3: 123.75 kW (increase of 33.75 kW)

[0267] Module 4: 118.5 kW (increase of 8.5 kW)

[0268] Module 5: 138 kW (decrease of 72 kW)

[0269] Module 6: 123 kW (increase of 83 kW)

[0270] Load migration volume of Module 5: ΔL PSO = 210 - 138 = 72 kW

[0271] Output fusion of double - layer optimization control

[0272] 1. Fusion of the two - layer control results (taking Module 5 as an example):

[0273] F timing_final = W fuzzy (t) × F timingfuzzy + W PSO (t) × F timingPSO = 0.233 × 0.98 + 0.767 × 1.0 = 0.228 + 0.767 = 0.995

[0274] Actual migration volume of Module 5:

[0275] ΔL_final = W fuzzy (t) × ΔL fuzzy + W PSO (t) × ΔL PSO = 0.233 × 51.85 + 0.767 × 72 = 12.08 + 55.22 = 67.3 kW

[0276] 2. Calculation of migration priority:

[0277] P migration_i = γ1L currenti + γ2Var(L i ) + γ3(1 - H i ) - P migration_1=0.6×120 + 0.3×15 + 0.1×(1 - 0.95)=72 + 4.5 + 0.005 = 76.505

[0278] P migration_2 =0.6×180 + 0.3×25 + 0.1×(1 - 0.90)=108 + 7.5 + 0.01 = 115.51

[0279] P migration_3 =0.6×90 + 0.3×20 + 0.1×(1 - 0.85)=54 + 6 + 0.015 = 60.015

[0280] P migration_4 =0.6×110 + 0.3×10 + 0.1×(1 - 0.98)=66 + 3 + 0.002 = 69.002

[0281] P migration_5 =0.6×210 + 0.3×35 + 0.1×(1 - 0.92)=126 + 10.5 + 0.008 = 136.508

[0282] P migration_6 =0.6×40 + 0.3×5 + 0.1×(1 - 1.00)=24 + 1.5 + 0 = 25.5

[0283] 3. Based on the migration priority, determine the source module and the target module:

[0284] Source module: Module 5 > Module 2 > Module 1 > Module 4 > Module 3 > Module 6

[0285] Target module: Module 6 < Module 3 < Module 4 < Module 1 < Module 2 < Module 5

[0286] 4. Determine the migration path and the load:

[0287] Migrate 67.3 kW from Module 5 and allocate it to Module 6 (40 kW) and Module 3 (27.3 kW)

[0288] Migrate 50.25 kW from Module 2 and allocate it to Module 3 (23 kW) and Module 4 (27.25 kW)

[0289] 5. Final result of the double - layer optimization control:

[0290] Module 1: 120 kW (unchanged)

[0291] Module 2: 180 - 50.25 = 129.75 kW

[0292] Module 3: 90 + 27.3 + 23 = 140.3 kW

[0293] Module 4: 110+27.25=137.25kW

[0294] Module 5: 210-67.3=142.7kW - Module 6: 40+40=80kW

[0295] 6. System parameters after double-layer optimization control adjustment:

[0296] Adjusted standard deviation: σ(L current _new)=22.06kW

[0297] Adjusted load imbalance: D imbalance _new=22.06 / 125=0.176

[0298] Summary and analysis:

[0299] 1. Load balancing: The two-layer optimization control system performs best in load balancing, with the load imbalance reduced from the initial 0.465 to 0.176, which is better than the dynamic balance system (0.311) and the predictive system (0.342). The static system is completely unbalanced, resulting in overload of a single module.

[0300] 2. Reasonable load distribution: The two-layer optimization control system takes into account the module characteristics and the overall system efficiency. The load of each module is relatively even and within the safe range. The maximum load rate is only 62.0%, which is much lower than other systems, providing the system with sufficient load margin to deal with emergencies.

[0301] 3. Intelligent decision-making ability: The two-layer optimization control system combines fuzzy rules and particle swarm optimization to more intelligently consider multiple factors such as load characteristics, ambient temperature and equipment health status, making decisions more comprehensive and reasonable. For example, considering the particularity of air-conditioning load under high temperature (35℃) in summer, the load distribution of module 5 is reasonably adjusted.

[0302] 4. Ability to adapt to dynamic changes: The two-layer optimization control system can dynamically adjust the control strategy according to the load change rate and system status. The adaptive weight function enables the system to flexibly switch the control strategy when the system status changes, and its adaptability far exceeds that of traditional systems.

[0303] 5. Computational resource balance: Although the computational complexity of the two-layer optimization control system is higher than that of the basic system, the system can select the appropriate computing intensity in different situations through the adaptive weight mechanism, balancing computing resources and optimization effects. In this case, due to the high system load imbalance (0.289 close to the threshold of 0.3), the particle swarm optimization result (weight 0.767) was used first, and a solution closer to the optimal solution was obtained.

[0304] 6. Rationality of migration strategy: The two-layer optimization control system comprehensively considers the current load, load fluctuation, and device health status through a multi-level load migration priority mechanism, making the migration decision more targeted. Although the total migration amount (117.55 kW) is greater than that of other systems, the migration effect is better and the system is more balanced.

[0305] Summary: The two-layer optimization control system realizes the intelligent management of multi-module electric meter loads by combining the particle swarm optimization system and fuzzy logic control technology, and has advantages compared with traditional basic systems, especially in terms of load balance, decision-making intelligence, and system adaptability. The system can dynamically adjust the control strategy according to the actual situation, ensuring both the optimization effect and computational efficiency, and providing an advanced management solution for complex and changing power load environments.

Claims

1. A multi-modular electric meter power management system for adaptive load management, characterized in that: The system includes: an electric meter module, a load monitoring unit, a data storage unit, and a double-layer optimization control unit, wherein the double-layer optimization control unit includes a particle swarm optimization layer and a fuzzy logic control layer, wherein: The load monitoring unit is used to collect the load level of each meter module in real time. , where i represents the number of the meter module; The data storage unit is used to store historical load data, user power usage patterns and environmental parameters; The particle swarm optimization layer is used to search for the optimal load distribution solution in the multidimensional decision space by iteratively updating the particle position vector X and the velocity vector V, where X represents the target load distribution ratio of each module and V represents the load distribution adjustment rate; The fuzzy logic control layer is used to take the load state and environmental parameters as input variables and generate load migration decisions through fuzzy rule reasoning, including the migration timing judgment function F timing And migration calculation function F amount ; The two-layer optimization control unit combines the search capability of the particle swarm optimization layer with the expert knowledge reasoning capability of the fuzzy logic control layer, and dynamically adjusts the weight ratio of the two-layer control output through the adaptive weight function W(t), thereby overcoming the disadvantages of the particle swarm system being prone to falling into local optimality and the disadvantages of the fuzzy logic control lacking optimization capability, and realizing dynamic balanced distribution of the load.

2. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The particle swarm optimization layer adopts an improved particle swarm optimization system, including: Adaptive inertia weight factor w(t), through the function Calculate, where t represents the current iteration number, T max represents the maximum number of iterations, w max and w min are the upper and lower limits of inertia weight respectively, α is the nonlinear adjustment factor; the particle position update formula is: ; Particle velocity update formula: ,in represents the best historical position of particle i, G best represents the best position found by the group, c1 and c2 are acceleration constants, r1 and r2 are random numbers in the interval [0,1]; The fitness function f(X) takes into account load balance, power quality and system efficiency, and is expressed as , where β i is the weight coefficient of each sub-goal, f balance (X) is the load balancing function, f quality (X) is the power quality sub-function, f efficiency (X) is the system efficiency sub-function.

3. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The fuzzy logic control layer includes: input fuzzification unit, the current load level of the meter module , Load change rate , System load imbalance D imbalance and ambient temperature T are converted into fuzzy sets; Fuzzy rule base, which contains a set of IF-THEN rules formed by expert knowledge, used to deal with the complexity and uncertainty of load migration decisions; Fuzzy reasoning mechanism, using Mamdani reasoning method to execute fuzzy rules; Output defuzzification unit, converting the fuzzy reasoning result into a clear load migration timing judgment value F timing and the calculated migration value F amount .

4. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The adaptive weight function W(t) dynamically adjusts the output weights of the particle swarm optimization layer and the fuzzy logic control layer according to the system operation status, specifically: When the system is in a stable state, W fuzzy (t)>W PSO (t), the output of the fuzzy logic control layer is preferentially adopted to reduce computational overhead; When the system load fluctuates greatly or the imbalance exceeds the preset threshold, W PSO (t)>W fuzzy (t), the output of the particle swarm optimization layer is preferentially used to obtain the optimal solution; W fuzzy (t)+W PSO (t)=1, ensuring weight normalization.

5. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The system also includes a load prediction unit, which predicts the load trend of future periods based on historical load data and calculates the future load using a recursive neural network model. , where k is the prediction step size, and the prediction results are used as the input of the particle swarm optimization layer and the fuzzy logic control layer to improve the prediction of decision making.

6. The multi-modular electric meter power management system for adaptive load management according to claim 2, characterized in that: The load balancing evaluation subfunction f in the fitness function balance (X) is calculated as the ratio of the standard deviation of each module's load level to the average load: , where σ(L current ) represents the standard deviation of the current load of each module, L average Indicates the average load level. The smaller the value of this sub-function is, the more balanced the load distribution is.

7. The multi-modular electric meter power management system for adaptive load management according to claim 3, characterized in that: The fuzzy rule base contains professional rules for different load scenarios, including: A ruleset for peak load scenarios for fast response and load spreading; A set of rules for valley load scenarios for equipment efficiency and energy conservation; A set of rules for fast-fluctuating load scenarios for system stability and buffering capabilities; For the rule set with seasonal changes, the load distribution strategy is dynamically adjusted according to the ambient temperature T and historical data for the same period.

8. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The system sets up a multi-level load migration priority mechanism, which comprehensively considers the current load level of the meter module. , Historical load fluctuation range and equipment health index H i Determine the source module and target module for load migration, and prioritize migrating load from high-load and healthy modules to low-load modules. The migration priority calculation formula is: , where γ1, γ2, and γ3 are weight coefficients.

9. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The system sets up a collaborative learning mechanism of particle swarm optimization system and fuzzy logic control, which is specifically manifested as follows: The search results of the particle swarm optimization system are used to dynamically adjust the membership function parameters of the fuzzy rules; The expert knowledge of fuzzy control is used to guide the initialization and mutation operations of the particle swarm; Through collaborative learning, the system can simultaneously overcome the disadvantages of slow convergence of particle swarm systems and difficult design of fuzzy logic rule bases.

10. The multi-modular electric meter power management system for adaptive load management according to claim 1, characterized in that: The system also includes an emergency response module. When an abnormality in the power grid or a sudden load change is detected, the system bypasses the conventional decision-making process and directly adopts a preset emergency load distribution strategy to ensure the robustness and safety of the system in extreme situations. The emergency response trigger conditions are: , where ε is the load imbalance threshold, η is the load change rate threshold, L average Indicates the average load level, is the load change rate.

Images