An electric energy meter load dynamic scheduling method, system, device and electric energy meter

Abstract

The present application relates to the field of data processing, more particularly, the present application relates to a kind of electric energy meter load dynamic scheduling method, system, device and electric energy meter, method includes: based on the data of the data processing equipment and photovoltaic power generation including each refrigeration equipment in the machine room of electric energy meter acquisition;Based on the data collected, using LSTM model to predict the demand load of refrigeration equipment;Based on the predicted demand load and the real-time load of refrigeration equipment, the adjustable load of refrigeration equipment is calculated;With adjustable load as boundary constraint, global optimization is carried out using improved PSO algorithm, and the optimal scheduling instruction is obtained, and is issued to execute.The present application takes the heat stability of machine room as premise, dynamically coordinates refrigeration load and photovoltaic output, and then while ensuring the safe and reliable operation of key equipment, realizes the maximization of energy efficiency and the optimization of energy cost, so as to improve the accuracy of load scheduling and the overall energy efficiency of system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing. More particularly, the present application relates to a power meter load dynamic scheduling method, system, device and power meter. BACKGROUND

[0002] In the current power system, the popularity of multi-element electrical equipment (such as refrigeration equipment, data processing equipment) and the large-scale access of distributed energy (such as photovoltaic) have put forward higher requirements for load fine scheduling. The existing load scheduling method is mostly based on simple threshold or single target optimization, which is difficult to adapt to complex scene requirements.

[0003] The PSO (Particle Swarm Optimization) algorithm has been tried to be used in the field of load scheduling because it is suitable for multi-variable and non-linear optimization problems. The algorithm can find a better load allocation scheme in a multi-device scene through group intelligence search. However, when solving multi-constrained and multi-target problems, due to the existence of device differences, energy fluctuations and power grid constraints and other complex factors, the search mechanism of the traditional PSO algorithm shows poor adaptability, resulting in limited load scheduling effect. SUMMARY

[0004] To solve the technical problem of limited load scheduling effect due to the limitations of the prior art, the present application provides a solution in the following aspects.

[0005] In a first aspect of the present application, a power meter load dynamic scheduling method is provided, comprising:

[0006] Based on the data collected in the power meter collection room including each refrigeration equipment, data processing equipment and photovoltaic power generation, the data includes the real-time load of the refrigeration equipment, the real-time temperature of the machine room, the real-time load of the data processing equipment and the real-time output of the photovoltaic power generation; the collected data is normalized;

[0007] Based on the collected data, an LSTM model is used to predict the demand load of the refrigeration equipment, wherein the training of the LSTM model uses an adaptive step sliding window to obtain training samples, and the loss function of the LSTM model fuses temperature constraints;

[0008] Based on the predicted demand load and the real-time load of the refrigeration equipment, the adjustable load of the refrigeration equipment is calculated;

[0009] The adjustable load is taken as a boundary constraint, and an improved PSO algorithm is used for global optimization to obtain optimal scheduling instructions and execute them; the improved PSO algorithm comprises a multi-objective fitness function and an adaptive inertia weight, the multi-objective fitness function fuses the calculated sensitive level, temperature deviation degree and photovoltaic output coefficient of the refrigeration equipment, and the adaptive inertia weight is dynamically adjusted based on the temperature deviation degree, photovoltaic output coefficient of the refrigeration equipment and real-time load of the data processing equipment.

[0010] Preferably, the acquisition process of the adaptive step length sliding window comprises:

[0011] The standard temperature range and temperature warning range in the machine room are acquired, the length of time that the temperature rises from the normal state to reach or exceed the temperature warning range is recorded, and the mean of all recorded lengths of time is taken as the window length of the sliding window;

[0012] The middle value of the standard temperature range is calculated, then the absolute value of the difference between the maximum real-time temperature in the window and the middle value is calculated, the absolute value is taken as the temperature deviation degree in the window, the product of the exponential term of the deviation degree and the window length is calculated, and the product result is rounded up to obtain the step length of the sliding window.

[0013] Preferably, before the adjustable load is calculated, if the temperature of the machine room is in the normal state, the calculation of the adjustable load is not performed; if the temperature of the machine room exceeds the temperature warning range, the calculation of the adjustable load is performed, and the calculation process comprises:

[0014] The difference between the predicted demand load and the real-time load of the refrigeration equipment is taken as the initial value of the adjustable load; the ratio of the real-time voltage of the refrigeration equipment to the rated voltage of the refrigeration equipment is taken as the voltage correction coefficient, and the product of the calculated voltage correction coefficient and the initial value of the adjustable load is taken as the final adjustable load.

[0015] Preferably, the acquisition process of the multi-objective fitness function comprises:

[0016] The ratio of the absolute value of the adjustable load of each refrigeration equipment to the system refrigeration equipment load range is calculated and taken as the first term;

[0017] The product of the calculated first term, the sensitive level of the refrigeration equipment and the exponential term of the temperature deviation degree is calculated, and the mean of the product is calculated by traversing all refrigeration equipment; the mean is multiplied by the reciprocal of the photovoltaic output coefficient to obtain the multi-objective fitness function.

[0018] Preferably, the ratio of the standard deviation of the historical load of each refrigeration device to the average of the standard deviations of the historical loads of all refrigeration devices is obtained as a first factor, the average of the standard deviations of the historical loads of all data processing devices in the machine room is obtained as a second factor, and the product of the first factor and the second factor is taken as the sensitivity level of the refrigeration device; and the ratio of the current photovoltaic real-time output to the historical maximum output is taken as the photovoltaic output coefficient.

[0019] Preferably, the adaptive inertia weight is obtained by the following process:

[0020] The adaptive inertia weight is the reciprocal of the exponential function of the product of the average of the exponential terms of the temperature deviation degrees of all refrigeration devices, the supplement of the photovoltaic output coefficient, and the load average of all data processing devices.

[0021] Preferably, in the improved PSO algorithm, for each refrigeration device, the corresponding adjustable load is taken as a reference, and the reference is scaled according to the calculated adaptive inertia weight to determine the search interval of the load scheduling value of each refrigeration device.

[0022] In the second aspect of the present application, a power meter load dynamic scheduling system is provided, comprising: a data acquisition module for synchronously acquiring power meter parameters and real-time photovoltaic power generation output; a demand load prediction module for predicting the demand load of refrigeration devices based on the acquired data using an LSTM model; an adjustable load calculation module for calculating the voltage-corrected adjustable load based on the predicted demand load and real-time load; and a load scheduling optimization module for calculating the load scheduling optimization using an improved PSO algorithm based on the adjustable load.

[0023] In the third aspect of the present application, a power meter load dynamic scheduling device is provided, comprising: a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the power meter load dynamic scheduling method described above.

[0024] In the fourth aspect of the present application, a power meter is provided, which integrates the power meter load dynamic scheduling system described above, and the system includes the power meter load dynamic scheduling device described above, for realizing the load dynamic scheduling function.

[0025] The present application has the following beneficial effects:

[0026] Firstly, by using an LSTM model fused with temperature constraints for load prediction and innovatively introducing a sliding window mechanism with adaptive step length, the system can dynamically capture the nonlinear and time-series correlation between the temperature of the machine room and the refrigeration load, significantly improving the accuracy of demand load prediction.

[0027] Then, based on the above high-precision predicted load, by calculating the difference with the real-time load and introducing a voltage correction coefficient, a more adjustable load is obtained which is more in line with the actual operation state of the equipment, effectively preventing equipment overload or regulation failure caused by voltage fluctuations or inaccurate evaluation, and defining a safe and feasible search space for the optimization algorithm.

[0028] Finally, taking the adjustable load as the boundary, an improved PSO algorithm is used for solving, the core of which is the multi-objective fitness function and the adaptive inertia weight obtained by improvement. The multi-objective fitness function ingeniously unifies the sensitive level, temperature deviation degree and photovoltaic output coefficient of the refrigeration equipment in one target, so that the scheduling instruction must prioritize the stable operation of the key equipment and the safety of the computer room temperature while pursuing energy saving. And the adaptive inertia weight dynamically adjusts the exploration and development ability of the algorithm according to the real-time state of the system (temperature, photovoltaic, data processing equipment load), ensuring that the optimal solution can be quickly and stably converged under complex working conditions, and the optimal scheduling instruction considering safety, economy and efficiency is generated. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a method flowchart of steps S1-S4 in the power meter load dynamic scheduling method of an embodiment of the present application.

[0030] Figure 2 is a schematic diagram of a power meter load dynamic scheduling system of an embodiment of the present application.

[0031] Figure 3 is a structural block diagram of a power meter load dynamic scheduling device of an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments.

[0033] The application scenario of the present application is: through real-time monitoring and collaborative scheduling of data processing equipment and refrigeration equipment by a power meter, dynamic and optimal distribution of power load is realized under the premise of ensuring that the computer room environment meets the process requirements, and clean energy such as on-site photovoltaic is preferentially utilized, finally achieving safe, efficient and green operation of the computer room.

[0034] Among them, the data processing equipment includes servers, computer clusters, etc., and the refrigeration equipment includes precision air conditioners and other temperature regulation units.

[0035] In one embodiment, referring to Figure 1 , a power meter load dynamic scheduling method is provided, which includes steps S1-S4, as follows:

[0036] S1: Collect data including each refrigeration equipment, each data processing equipment, and photovoltaic power generation in the power meter collection room based on the power meter.

[0037] At a frequency of once per minute, the following key data is collected by means of the power meter:

[0038] The real-time load of each refrigeration equipment in the room, the voltage in the power system, the real-time temperature of the room, the real-time load of each data processing equipment, and the real-time output PV of photovoltaic. Further, all the collected data is subjected to conventional normalization processing.

[0039] After the above data collection and preprocessing steps, the data obtained is more standardized and easy to process, which can provide good input for multi-constraint and multi-objective optimization algorithms, helping the algorithm to find the optimal scheduling scheme in a complex scenario, and realizing efficient use of energy and safe operation of equipment.

[0040] S2: Based on the collected data, use the LSTM model to predict the demand load of the refrigeration equipment.

[0041] There are refrigeration equipment and data processing equipment in the room, and the operation of the data processing equipment will generate heat, causing the temperature in the room to rise. If the temperature is too high, it may affect the performance and life of the data processing equipment. By predicting the demand load of the refrigeration equipment, i.e. predicting the load required by the refrigeration equipment to maintain the temperature in the room within the standard range, it can ensure that the refrigeration equipment operates reasonably according to the prediction results, avoid the room temperature being too high due to insufficient refrigeration, and ensure the stable operation of the data processing equipment in the room.

[0042] Use historical data (obtained according to the above S1 data collection operation) to train the LSTM model (Long Short-Term Memory, Long Short-Term Memory Network), learn how to maintain the temperature in the room stable, and output the load value required by the refrigeration equipment.

[0043] First, the model input comes from historical time series data, including three feature dimensions: normalized real-time load of refrigeration equipment, normalized real-time temperature of the room, and normalized real-time load of data processing equipment.

[0044] Then, considering the complexity of the room environment, the demand load of the refrigeration equipment and the temperature fluctuate over time, for example, the data processing equipment runs at different intensities in different business periods, producing different amounts of heat, causing the room temperature and the demand load of the refrigeration equipment to change accordingly.

[0045] Therefore, the sliding window mechanism with adaptive step length is used when obtaining the training samples, which can adjust the window step length according to the real-time temperature, and when the temperature changes dramatically and deviates from the standard range, the step length is reduced, the data is intercepted more frequently, and the dynamic changes are captured in time, so that the model can learn more detailed load change patterns.

[0046] For example, by consulting the specification of all key equipment in the machine room, the working temperature requirement is found, and within the allowed range of all equipment, a final and unified standard temperature range is determined in combination with industry recommendations and its own policy. Then, according to the standard temperature range, the temperature warning range is set, that is, by shrinking 10% of the temperature difference to the middle of the upper and lower limits of the standard temperature range, a relatively stricter and more easily triggered temperature interval for early warning is formulated. For example, if the standard temperature range is [20℃, 30℃], then the temperature warning range is [21℃, 29℃].

[0047] The above standard temperature range is an ideal temperature interval set to ensure the normal operation and stable performance of the equipment in the machine room, and the temperature warning range is a sub-range of the standard temperature range, which is relatively stricter and is used to warn in advance that the temperature may have an adverse effect on the equipment.

[0048] By observing the temperature data input into the above model, the length of time that the temperature rises from the normal state (within the standard temperature range) to reach or exceed the temperature warning range is recorded, because in the actual machine room scenario, the temperature rising above or exceeding the upper limit is more common and needs attention.

[0049] For example, within a certain period of time, the temperature rises from 22℃ (normal) to 29.5℃ (exceeding the upper limit of the warning range), and if this process takes 30 minutes, then the 30 minutes is a recorded time interval. All such recorded time intervals are summarized and the average is calculated. For example, 5 time intervals are recorded, which are 30 minutes, 40 minutes, 35 minutes, 25 minutes and 45 minutes, and the average is 35 minutes, which is the window length of the sliding window.

[0050] If the calculated average is not an integer, it needs to be rounded to the nearest integer.

[0051] Further, the step length of the window sliding is dynamically adjusted according to the temperature deviation within the window length. The specific operation is as follows:

[0052] The middle value of the standard temperature range is calculated, and then the absolute value of the difference between the maximum real-time temperature in the window and the middle value is calculated, which is taken as the temperature deviation within the window. The product of the exponential term of the deviation and the window length is calculated, and the result of the product is rounded up to obtain the step length of the sliding window.

[0053] The calculation formula of the step length is as follows:

[0054]

[0055] In the formula, is the step length of the sliding window, is the temperature deviation degree in the window, is the window length.

[0056] In summary, the calculation method of the window length of the sliding window and the sliding step length at each time is obtained, an adaptive step length sliding window mechanism is obtained, and the data input into the above model is intercepted to obtain the training sample and the label corresponding to the training sample. The label is the average of the actual load of the refrigeration equipment in the time steps after the end of the window.

[0057] Secondly, the loss function required for model training is introduced to improve the performance of the model in maintaining the stability of the environment temperature.

[0058] That is, the mean square error of all existing samples and the sensitivity index of the load scheduling of the sample based on the temperature calculation are fused to obtain the loss function required for model training.

[0059] For example, the sum of squares of the deviation degree of all temperature values in the window (i.e., the absolute value of the difference between each temperature value in the window and the middle value of the standard temperature range) is calculated, and then the hyperbolic tangent function is transformed to obtain the sensitivity of the load scheduling of the sample. Through this transformation, the temperature deviation degree is converted into a sensitivity index in a specific interval, so that when the actual temperature is closer to or exceeds the standard temperature threshold (i.e., the deviation degree is large), the index value is larger, thereby highlighting the loss weight of the high load scheduling sensitivity sample.

[0060] The above loss function is expressed by the relationship as follows:

[0061]

[0062] In the formula, is the predicted demand load of the sample , and is the mean square error of the real load , and is the sensitivity of the load scheduling of the sample , and is the total number of training samples, is the hyperbolic tangent function.

[0063] ​Through the above operation, an LSTM model for predicting the demand load of the refrigeration equipment is obtained. The model uses an adaptive sliding window to process real-time data and is trained using a comprehensive loss function, and can accurately predict the load required to maintain the temperature. This provides a key input for the intelligent load scheduling system, ensuring that the temperature in the machine room is always controlled within the standard range.

[0064] Finally, according to the trained LSTM model above, input the real-time data of the past time steps, and output the predicted value of the average load required by the refrigeration equipment in the future time steps, i.e. the predicted demand load of the refrigeration equipment.

[0065] It should be noted that the operation of S2 above is performed for a single refrigeration equipment.

[0066] S3: Based on the predicted demand load and the real-time load of the refrigeration equipment, calculate the adjustable load.

[0067] Although the predicted demand load of the refrigeration equipment has been obtained in S2 above, it needs to be considered that the reduction or increase of the load will directly affect the heat generation of the equipment in the machine room, and thus affect the temperature in the machine room. If load scheduling is only based on predicted load without considering current temperature conditions, it may lead to further loss of control of temperature. For example, if the predicted load shows that the load needs to be reduced, but if the current temperature is close to the warning value, blindly reducing the load may cause the refrigeration equipment to run insufficiently, resulting in a continued rise in temperature. Therefore, taking the temperature exceeding the warning value as the trigger condition, the mutual influence of load and temperature can be considered comprehensively.

[0068] When the temperature in the machine room exceeds the temperature warning range, load scheduling is triggered according to the prediction result of S2 above, and then the calculation of the adjustable load of the refrigeration equipment is performed. The calculation process of the adjustable load is as follows:

[0069] The difference between the predicted demand load and the real-time load is used as the initial value of the adjustable load, and the initial value of the adjustable load is corrected in combination with the power safety constraints of the refrigeration equipment to avoid excessive adjustment of the equipment load and cause power danger when the voltage deviates from the rated voltage.

[0070] Among them, the ratio of the real-time voltage of the refrigeration equipment to the rated voltage of the refrigeration equipment is used as the voltage correction coefficient, and the voltage correction coefficient obtained by calculation is multiplied by the initial value of the adjustable load to obtain the final adjustable load.

[0071] S4: Using the improved PSO algorithm for global optimization with adjustable load as boundary constraint, the optimal scheduling instruction is obtained and executed.

[0072] Since the S3 calculation is the theoretical adjustable load of each refrigeration equipment, in order to avoid all the equipment adjusting at the same time, it is necessary to reasonably allocate the adjustment task among multiple equipment to find the overall optimal scheduling scheme, so that the load scheduling effect is optimal.

[0073] Considering the priority of the safety of the machine room and the energy orientation, by analyzing the temperature deviation of the machine room, the sensitivity of the equipment and the photovoltaic output, the three constraint conditions are integrated into the fitness function and the inertia weight of the PSO algorithm, so that the scheduling algorithm can dynamically adjust the search strategy according to the actual situation, which not only ensures safety, but also improves energy utilization.

[0074] The above improvement of the PSO algorithm includes obtaining a multi-objective fitness function and an adaptive inertia weight.

[0075] The step of obtaining the multi-objective fitness function is as follows:

[0076] Firstly, the ratio of the historical load standard deviation of each refrigeration equipment to the average of the historical load standard deviation of all refrigeration equipment is obtained as the first factor, and the average of the historical load standard deviation of all data processing equipment in the machine room is obtained as the second factor, and the product of the first factor and the second factor is taken as the sensitivity level of the refrigeration equipment.

[0077] Then, the absolute value of the difference between the real-time temperature and the center value of the standard temperature range is converted into an exponential term (in the form of ), and the exponential term is taken as the temperature deviation degree of the refrigeration equipment.

[0078] Secondly, the ratio of the current photovoltaic real-time output to the historical maximum output is taken as the photovoltaic output coefficient.

[0079] Finally, the absolute value of the adjustable load of each refrigeration equipment and the ratio of the system refrigeration equipment load range are calculated, and the product of the three, the sensitivity level of the refrigeration equipment and the exponential term of the temperature deviation degree, is taken as the first item, and the average of the product of all refrigeration equipment is calculated; the reciprocal of the photovoltaic output coefficient is multiplied by the average to obtain the multi-objective fitness function.

[0080] Exemplarily, the multi-objective fitness function is expressed by a relational expression as follows:

[0081]

[0082] In the formula, is the multi-objective fitness function, is the absolute value of the adjustable load of the first refrigeration equipment, is the ratio of the absolute value of the adjustable load of the first refrigeration equipment to the system refrigeration equipment load range (i.e. the difference between the maximum value and the minimum value of the load range), is the reciprocal of the photovoltaic output coefficient. sensitivity level of the refrigeration equipment, is the total number of refrigeration equipment, is an index of the temperature deviation of the refrigeration equipment, is the total number of refrigeration equipment, is the photovoltaic output coefficient, is a constant to prevent the denominator from being zero. Wherein, the first part reflects the comprehensive cost or influence of refrigeration equipment load adjustment, and the second part introduces the influence of photovoltaic output, when the photovoltaic output is high, is reduced, encouraging load adjustment when photovoltaic power generation is sufficient to reduce dependence on the power grid.

[0083] By minimizing the multi-objective fitness function value, a set of optimal load scheduling strategies can be found, that is, the load setting of each refrigeration equipment is determined, so as to balance multiple objectives: making full use of the adjustment capacity of the equipment, maintaining temperature comfort, using renewable energy and reducing operating costs.

[0084] The adaptive inertia weight is obtained as follows:

[0085] The reciprocal of the exponential function of the product of the mean of the index of the temperature deviation of all refrigeration equipment, the complement of the photovoltaic output coefficient and the mean of the load of all data processing equipment is taken as the adaptive inertia weight.

[0086] Exemplarily, the adaptive inertia weight satisfies the relationship:

[0087]

[0088] In the formula, is the adaptive inertia weight, is the mean of the index of the temperature deviation of all refrigeration equipment, is the photovoltaic output coefficient, is the mean of the load of all data processing equipment in the machine room, is the exponential function with the natural number e as the base. When is large, is low, is large, small, indicating that the refrigeration equipment is in a tense state and needs more active adjustment; otherwise, the refrigeration equipment is in good condition and allows more conservative exploration.

[0089] According to the above operation, the core components of the improved PSO algorithm are obtained, and then the final operation is performed. The specific process is as follows:

[0090] In the improved PSO algorithm of the present application, the particle is a The search interval of each dimension is dynamically determined based on the adaptive inertia weight calculated above and the adjustable load of the refrigeration equipment. Each dimension represents the search interval of the adjustable load of the refrigeration equipment. The adjustable load of the refrigeration equipment.

[0091] For the refrigeration equipment that needs to be scheduled, the search interval of the adjustable load of each refrigeration equipment is determined by scaling the corresponding adjustable load according to the calculated adaptive inertia weight. The search interval of each dimension satisfies:

[0092] When the adjustable load of the i-th dimension is greater than 0, the value range is: .

[0093] When the adjustable load of the i-th dimension is less than 0, the value range is: .

[0094] In summary, the multi-objective fitness function is used as an evaluation criterion for the quality of particles (i.e. scheduling schemes). By minimizing the function value, the particle swarm is directly guided to search in the direction of the lowest comprehensive cost, ensuring that the final solution can simultaneously consider equipment adjustment fairness, data center temperature safety, and efficient use of photovoltaic energy. The adaptive inertia weight dynamically controls the search behavior and search interval of the particle swarm. The weight is adaptively adjusted according to the real-time state of the system (average temperature deviation, photovoltaic output, and data processing equipment load), thereby achieving a balance between "global exploration" and "local fine search". This not only improves the convergence speed and accuracy of the algorithm, but also makes the generated scheduling instructions more robust and able to respond to environmental changes in real time.

[0095] Finally, the improved PSO algorithm based on the above mechanism efficiently searches for the globally optimal refrigeration equipment load scheduling instruction within the solution space constrained by the dynamically determined search interval. Then, according to the optimal load scheduling instruction, a series of specific refrigeration equipment load instructions are generated, which are directly issued to the smart energy meters connected to each refrigeration equipment through the energy management system. The smart energy meters accurately adjust the refrigeration equipment to the target range according to the received instructions.

[0096] In one embodiment, referring to Figure 2 , a power meter load dynamic scheduling system is provided, comprising:

[0097] A data acquisition module is used to synchronously acquire power meter parameters and real-time photovoltaic power generation output.

[0098] A demand load prediction module is used to predict the demand load of the refrigeration equipment based on the acquired data using an LSTM model. ​​

[0099] an adjustable load calculation module configured to calculate a voltage-corrected adjustable load based on the predicted demand load and the real-time load;

[0100] a load scheduling optimization module configured to calculate a load scheduling optimization based on the adjustable load using an improved PSO algorithm.

[0101] In one embodiment, referring to Figure 3 , there is provided an electric energy meter load dynamic scheduling device, comprising:

[0102] a processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement the electric energy meter load dynamic scheduling method described above.

[0103] In one embodiment, there is provided an electric energy meter integrating the electric energy meter load dynamic scheduling system described above, and the system includes the electric energy meter load dynamic scheduling device described above for implementing the load dynamic scheduling function.

[0104] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the inventive concept, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for dynamic load scheduling of electricity meters, characterized in that, include: Data is collected from the electricity meter, including data from various cooling devices, data processing equipment, and photovoltaic power generation within the computer room. This data includes the real-time load and voltage of the cooling devices, the real-time temperature of the computer room, the real-time load of the data processing equipment, and the real-time output of the photovoltaic power generation. The collected data is then normalized. Based on the collected data, an LSTM model is used to predict the demand load of the refrigeration equipment. The LSTM model is trained using an adaptive step-size sliding window to obtain training samples, and the loss function of the LSTM model incorporates temperature constraints. The process of obtaining an adaptive step-size sliding window includes: Obtain the standard temperature range and temperature warning range in the computer room, record the time it takes for the temperature to rise from normal to reach or exceed the temperature warning range, and use the average of all recorded time lengths as the window length of the sliding window. Calculate the median value of the standard temperature range, then calculate the absolute value of the difference between the real-time maximum temperature in the window and the median value. Use this absolute value of the difference as the degree of temperature deviation within the window. Calculate the product of the exponent of this degree of deviation and the window length, and round the product up to obtain the step size of the sliding window. Based on the predicted demand load and the real-time load of the refrigeration equipment, calculate the adjustable load of the refrigeration equipment; Using adjustable load as boundary constraint, an improved PSO algorithm is used for global optimization to obtain the optimal scheduling instruction, which is then issued for execution. The improved PSO algorithm includes a multi-objective fitness function and an adaptive inertial weight. The multi-objective fitness function integrates the sensitivity level, temperature deviation, and photovoltaic output coefficient of the refrigeration equipment. The adaptive inertial weight is based on the temperature deviation of the refrigeration equipment, the photovoltaic output coefficient, and the real-time load dynamic adjustment of the data processing equipment. The ratio of the historical load standard deviation of each cooling device to the mean of the historical load standard deviation of all cooling devices is taken as the first factor. The mean of the historical load standard deviation of all data processing devices in the computer room is taken as the second factor. The product of the first factor and the second factor is taken as the sensitivity level of the cooling device. The ratio of the current real-time output of photovoltaic power to the historical maximum output is taken as the photovoltaic output coefficient.

2. The method for dynamic load scheduling of an electricity meter according to claim 1, characterized in that, If the computer room temperature is within normal limits before calculating the adjustable load, the adjustable load calculation will not be performed; if the computer room temperature exceeds the temperature warning range, the adjustable load calculation will be performed, and the calculation process includes: The difference between the predicted demand load and the real-time load of the refrigeration equipment is used as the initial value of the adjustable load; the ratio of the real-time voltage of the refrigeration equipment to the rated voltage of the refrigeration equipment is used as the voltage correction factor; the calculated voltage correction factor is multiplied by the initial value of the adjustable load to obtain the final adjustable load.

3. The method for dynamic load scheduling of an electricity meter according to claim 2, characterized in that, The process of obtaining the multi-objective fitness function includes: Calculate the ratio of the absolute value of the adjustable load of each refrigeration unit to the load range of the system's refrigeration units, and use this as the first item; The product of the first term, the sensitivity level of the refrigeration equipment, and the exponential term of the temperature deviation is calculated, and the mean of the product is calculated by iterating through all refrigeration equipment. This mean is then multiplied by the reciprocal of the photovoltaic output coefficient to obtain the multi-objective fitness function.

4. The method for dynamic load scheduling of an electricity meter according to claim 3, characterized in that, The process of obtaining the adaptive inertia weight includes: The negative of the exponential function of the product of the mean of the exponential term of the temperature deviation of all refrigeration equipment, the complement of the photovoltaic output coefficient, and the mean load of all data processing equipment is used as the adaptive inertia weight.

5. The method for dynamic load scheduling of an electricity meter according to claim 4, characterized in that, In the improved PSO algorithm, for each refrigeration unit, the corresponding adjustable load is used as a reference, and the reference is scaled according to the calculated adaptive inertia weight to determine the search range of the load scheduling value for each refrigeration unit.

6. A dynamic load scheduling system for electricity meters, used to implement the dynamic load scheduling method for electricity meters according to any one of claims 1-5, characterized in that, include: The data acquisition module is used to synchronously collect the parameters of the electricity meter and the real-time output of photovoltaic power generation; The demand load forecasting module is used to predict the demand load of refrigeration equipment based on the collected data and using an LSTM model. The adjustable load calculation module is used to calculate the voltage-corrected adjustable load based on the predicted demand load and the real-time load. The load scheduling optimization module is used to calculate the optimal load scheduling based on the adjustable load using an improved PSO algorithm.

7. A dynamic load dispatching device for electricity meters, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the dynamic load scheduling method for electricity meters according to any one of claims 1-5.

8. An electricity meter, characterized in that, The system integrates the dynamic load scheduling system for electricity meters as described in claim 6, and the system includes the dynamic load scheduling device for electricity meters as described in claim 7, for realizing the dynamic load scheduling function.

Images