Rail transit-oriented air conditioner intelligent operation and maintenance method and system

Abstract

The present application relates to the field of rail transit air conditioning management, and particularly relates to a rail transit-oriented air conditioning intelligent operation and maintenance method and system, which comprises the following steps: collecting air conditioner power, passenger flow density and temperature of all divided areas in a carriage in real time; constructing an LSTM model, wherein the input of the LSTM model is time series of air conditioner power, passenger flow density and temperature, and the output is predicted temperature drop time series; adjusting air conditioner power of each divided area; and managing air conditioners of each divided area to operate at the adjusted power. The present application can realize real-time sensing of heat load mutation in the carriage caused by passenger flow, dynamically adjust air conditioner power of each divided area, make air conditioner power distribution closely combined with real-time demand, improve passenger comfort, make high passenger flow density areas obtain sufficient refrigeration power, low passenger flow density areas avoid energy waste, and improve overall resource utilization efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail transit air conditioning management. In particular, it relates to an intelligent operation and maintenance method and system for rail transit air conditioning. BACKGROUND

[0002] As an efficient and convenient urban public transportation mode, the comfort of the operating environment of rail transit is crucial to passenger experience. The air conditioning system, as a key device to ensure the comfort of the vehicle environment, its operation efficiency and maintenance management are directly related to the comfort of passengers and operating costs. However, the current rail transit air conditioning system generally adopts a group control strategy based on fixed thresholds or an offline model prediction method. These methods have significant shortcomings in dynamic response, model generalization ability and resource allocation.

[0003] For example, existing systems usually adopt a fixed power allocation strategy, ignoring the differences in refrigeration benefits between regions. High-density areas may not be cooled enough, while low-density areas may waste energy, resulting in low energy utilization efficiency, increasing operating costs, and affecting passenger comfort. SUMMARY

[0004] To solve the above technical problem of low energy utilization efficiency of the operation and maintenance method, the present application provides a solution in the following aspects.

[0005] In a first aspect, the present application discloses an intelligent operation and maintenance method for rail transit air conditioning, comprising:

[0006] Real-time collection of air conditioning power, passenger flow density and temperature of all divided regions in the vehicle;

[0007] Constructing an LSTM model, the input of the LSTM model is the time series of air conditioning power, passenger flow density and temperature, and the output is the predicted temperature drop time series. The air conditioning power of each divided region is adjusted, and the air conditioning of each divided region is operated at the adjusted power;

[0008] The process of the allocation comprises: for a single division area, predicting a predicted temperature drop time sequence when the air conditioner power is increased by a preset power and a predicted temperature drop time sequence when the air conditioner power is reduced by the preset power by using an LSTM model, calculating a sensitive value of the corresponding division area to the change of the air conditioner power according to the two predicted temperature drop time sequences, and marking the division area as a stable area or a non-stable area according to the sensitive value; setting a to-be-allocated pool, extracting a first power in the current air conditioner power in each stable area, obtaining a first adjusted power after the first power is weakened, setting an extraction condition according to the predicted temperature drop time sequence under the first adjusted power and the predicted temperature drop time sequence under the current air conditioner power, and in response to the extraction condition being met, putting the first power into the to-be-allocated pool; in each non-stable area, extracting a second power in the to-be-allocated pool, adding the second power to the current air conditioner power of the non-stable area to obtain a second adjusted power, setting an update condition according to the predicted temperature drop time sequence under the second adjusted power and the predicted temperature drop time sequence under the current air conditioner power, and in response to the update condition being met, updating the current air conditioner power to the second adjusted power.

[0009] The application can realize real-time sensing of thermal load mutation caused by passenger flow in the carriage, dynamically adjusts the air conditioner power of each division area, so that the air conditioner power allocation can be closely combined with real-time demand, improves the passenger comfort, at the same time, makes the high passenger flow density area obtain sufficient refrigeration power, the low density passenger flow area avoids energy waste, and improves the overall resource utilization efficiency.

[0010] Preferably, before the air conditioner power of each division area is allocated, the application further comprises: for a single division area, calculating the confidence of the output result of the LSTM model, when the confidence is less than a preset correction threshold continuously for a set number of times, correcting the dynamic parameters of the LSTM model, and when the number of correction times is greater than a preset number of times within a preset time, generating a maintenance work order.

[0011] Through the linkage of the confidence and the cumulative correction times, the parameter correction or the generation of the maintenance work order is automatically triggered, the decision-making closed loop from algorithm optimization to hardware maintenance is realized, the equipment failure response time is greatly shortened, the operation and maintenance efficiency is improved, the air conditioner system maintains good performance in long-term operation, and the maintenance cost is reduced.

[0012] Preferably, the calculation process of the confidence comprises:

[0013] In the predicted temperature drop time sequence output by the LSTM model, a predicted temperature drop value corresponding to the current time is selected, the error between the predicted temperature drop value and the actual temperature drop value at the current time is calculated, a protection threshold is set, the ratio of the absolute value of the error to the maximum of the protection threshold and the predicted temperature drop value is calculated, and the difference between 1 and the ratio is taken as the confidence.

[0014] By fusing the online learning mechanism through the LSTM model, the temperature drop prediction error is obviously reduced compared with the traditional model, and high confidence can still be maintained in the equipment performance degradation scenario.

[0015] Preferably, the obtaining process of the sensitive value comprises: selecting a first temperature drop stable value from the predicted temperature drop time sequence when the air conditioner power increases by a preset power, and selecting a second temperature drop stable value from the predicted temperature drop time sequence when the air conditioner power decreases by the preset power; and calculating a central difference gradient based on the first temperature drop stable value and the second temperature drop stable value to obtain the sensitive value.

[0016] Preferably, the marking of the divided region as a stable region or a non-stable region according to the sensitive value specifically comprises: when the sensitive value is greater than or equal to a preset fixed value, marking the corresponding divided region as a non-stable region; and when the sensitive value is less than the preset fixed value, marking the corresponding divided region as a stable region.

[0017] Preferably, before the air conditioner power of the non-stable region is adjusted, the method further comprises: arranging the non-stable regions in descending order according to the size of the sensitive value, and adjusting the air conditioner power of the non-stable regions in turn according to the arrangement order.

[0018] Preferably, the setting method of the extraction condition is: selecting a first stable value from the predicted temperature drop time sequence under the first adjustment power and the predicted temperature drop time sequence under the current air conditioner power respectively, calculating the absolute value of the difference between the two first stable values to obtain a first temperature drop change value, and when the first temperature drop change value is less than a preset value, the extraction condition is met.

[0019] Preferably, the setting method of the update condition is: selecting a second stable value from the predicted temperature drop time sequence under the second adjustment power and the predicted temperature drop time sequence under the current air conditioner power respectively, calculating the absolute value of the difference between the two second stable values to obtain a second temperature drop change value, and when the second temperature drop change value is greater than a preset value, the update condition is met.

[0020] Preferably, the obtaining process of the second power comprises: calculating the ratio of the current sensitive value of the corresponding non-stable region to the sum of the sensitive values of all divided regions, and calculating the product of the ratio and the total power in the to-be-allocated pool to obtain the second power.

[0021] In a second aspect, the application further discloses an air conditioner intelligent operation and maintenance system for rail transit, which comprises a processor and a memory, and the memory stores computer program instructions.

[0022] The application has the following effects:

[0023] 1、The application can realize real-time sensing of heat load mutation caused by passenger flow in the carriage, dynamically adjusts the air conditioning power of each divided area, so that the air conditioning power distribution can be closely combined with the real-time demand, improves the passenger comfort, at the same time, makes the high passenger flow density area obtain sufficient refrigeration power, avoids energy waste in the low density passenger flow area, and improves the overall resource utilization efficiency.

[0024] 2、The application fuses an online learning mechanism through an LSTM model, the temperature drop prediction error is obviously reduced compared with a traditional model, and high confidence can be maintained in the equipment performance degradation scene, the adaptability of the model to the dynamic change of the air conditioning system is improved, the model can accurately predict the temperature drop under different operating conditions, thereby optimizing the management of the air conditioning system operation. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a method flowchart of steps S1-S3 in an air conditioning intelligent operation and maintenance method for rail transit.

[0026] Figure 2 is a method flowchart of steps S30-S32 in an air conditioning intelligent operation and maintenance method for rail transit. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application.

[0028] The specific embodiments of the application will be described in detail below with reference to the drawings.

[0029] Referring to Figure 1 An air conditioning intelligent operation and maintenance method for rail transit includes steps S1-S7, and specifically as follows:

[0030] S1: Real-time collection of air conditioning power, passenger flow density and temperature of all divided areas in the carriage.

[0031] First, the subway carriage is reasonably divided, the total number of divided areas is determined according to the number of air conditioners in the subway carriage, the area of each divided area is ensured to match the effective coverage range of the air conditioner, and the division of each divided area is mainly based on actual factors such as the position of the air conditioner outlet, the seat layout and the passenger flow characteristics, especially the area near the door needs to be divided separately, because the passenger flow density changes most significantly. For example, if the number of air conditioners in the subway carriage is 10, then 10 divided areas are needed, and each air conditioner controls a divided area, which can be divided according to the position of the air conditioner outlet, the passenger flow characteristics or whether it is the area near the door and other factors.

[0032] The data acquisition system consists of a wide-angle camera installed on the top of the carriage and a high-precision digital temperature sensor. The camera is used to count the number of people in each divided area in real time, and the image recognition algorithm is used to calculate the crowd density. The sampling frequency is associated with the subway timetable. Since the positions of the people in each area are relatively fixed after the doors are closed for a period of time, the data collected at this time can accurately reflect the crowd density in each area. Therefore, data collection is started after a certain period of time (e.g., 30 seconds) at each station to ensure stable passenger distribution. Air conditioning power and crowd density are collected simultaneously.

[0033] The high-precision digital temperature sensor is arranged at the center of each divided area to record the temperature value of the divided area at a set time interval (e.g., 10 seconds) to form a temperature time sequence. After preprocessing the temperature time sequence, the first-order difference is calculated to generate a temperature drop sequence.

[0034] Finally, the data collected at each time step for each divided area is obtained, including the real-time air conditioning power, crowd density at the collection time point, the temperature time sequence formed by the temperature values collected between the adjacent two collection time points, and the corresponding temperature drop sequence.

[0035] The long short-term memory network is used as the prediction framework, and its gating mechanism is particularly suitable for handling the time delay effect and thermal inertia problem of the air conditioning system. This network can effectively capture the long-term dependence relationship in time series data, thereby improving the accuracy of prediction. For each divided area, an LSTM (Long Short-Term Memory) model is trained, and the following features are used as input during training: air conditioning power, crowd density, and temperature time sequence of the current divided area. The target output is a predicted temperature drop time sequence consisting of temperature drop values at future N time points.

[0036] When the LSTM model is called to verify the cooling effect of the current real-time air conditioning power of the corresponding divided area, the following features are input into the LSTM model: real-time air conditioning power, crowd density, and initial temperature of the current divided area. The LSTM model outputs a predicted temperature drop time sequence containing temperature drop values at future N time points. The predicted temperature drop time sequence reflects the change of temperature over time under the current air conditioning power and crowd density conditions, i.e., the LSTM model predicts the temperature change curve over time under the current air conditioning power and crowd density.

[0037] S2: For a single divided area, calculate the confidence of the LSTM model output result. When the confidence meets the first condition, modify the dynamic parameters of the LSTM model. When the number of modifications meets the second condition, generate a maintenance warning.

[0038] To solve the prediction deviation caused by the performance attenuation of the air conditioner, a double-layer parameter updating strategy is designed. The LSTM model forms a data pair with each input and output, and a sliding window queue with a length of 500 is defined to store the latest input and output data pairs of the LSTM model.

[0039] For a single divided area, after obtaining the latest collected air conditioner power, crowd density and temperature time series each time, the corresponding LSTM model is input, and the predicted temperature drop time series is output. Then the confidence of the LSTM model output result is calculated, and the confidence calculation process includes: selecting the predicted temperature drop value corresponding to the current time in the predicted temperature drop time series, calculating the error between the predicted temperature drop value and the actual temperature drop value at the current time; setting a protection threshold, calculating the ratio of the absolute value of the error to the maximum of the protection threshold and the predicted temperature drop value, and taking the difference between 1 and the ratio as the confidence.

[0040] The confidence is calculated by the following formula:

[0041]

[0042] In the formula, confidence represents the confidence; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; act represents the actual temperature drop value at the current time; error represents the absolute error between the predicted temperature drop value and the actual temperature drop value, which is always positive; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; is a set protection threshold; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series; pred represents the predicted temperature drop value at the current time in the predicted temperature drop time series;

[0043] When the confidence meets the first condition, i.e. the calculated confidence is less than the preset correction threshold (such as 0.8) for a set number of times (such as 5 times), the parameter correction process is automatically triggered. In this embodiment, the random gradient descent method is used to correct the dynamic parameters (mainly including output gate bias and cell state related weights) of the loss function in the LSTM model. The random gradient descent method is a prior art and will not be described here. The loss function in the LSTM model is a composite loss function combining the prediction error and the regularization term, which is as follows:

[0044]

[0045] In the formula, represents a loss function; represents the number of data pairs in the sliding window; represents the prediction temperature drop time series of the LSTM model of the th division area, which depends on static parameters and dynamic parameters ; represents the actual temperature drop sequence of the th division area; represents the 2-norm of the prediction error; represents the adaptive constraint strength coefficient.

[0046]

[0047] wherein, represents the value of the dynamic parameter used by the LSTM model in the current iteration step; represents the initial dynamic parameter value; represents a small constant used to ensure that the denominator is not zero. When the parameter offset is small ( ), , the parameter is forced to remain stable; when the offset is significantly increased (performance decay intensifies), , the constraint is relaxed to quickly adapt to changes.

[0048] is a regularization term. Since performance decay is usually a local and slow change, the parameter will not deviate too much from the initial value. The regularization term ensures that the correction is only for the part of the deviation caused by performance decay, rather than completely changing the model's basic understanding of the air conditioning system. Secondly, the amount of sliding window data (500) is much smaller than the initial training data. Without regularization, the model is easily over-fitted to the noise or short-term fluctuations of the 500 samples, losing the generalization ability. Therefore, the regularization term prevents the model from over-fitting the data in the sliding window and maintains the model's generalization ability.

[0049] When the number of corrections of the dynamic parameter meets the second condition, i.e., the number of corrections in a preset time is greater than a preset number (such as the monthly cumulative number of corrections of a certain division area exceeds 50), a maintenance warning is generated, realizing closed-loop management from algorithm adjustment to hardware maintenance.

[0050] The LSTM model in this embodiment integrates an online learning mechanism, which significantly reduces the temperature drop prediction error compared to traditional models, and the temperature drop prediction error still has high confidence in the device performance decay scenario. Through the linkage of confidence and cumulative correction number, parameter correction or maintenance work order is automatically triggered, realizing the decision-making closed loop from algorithm optimization to hardware maintenance, and greatly shortening the device failure response time.

[0051] S3: Adjust the air conditioning power of each divided area, and manage the air conditioner of each divided area to run at the adjusted power.

[0052] After collecting data each time, the air conditioning power is adjusted according to the current situation, aiming to dynamically optimize the air conditioning power distribution of each divided area. By quantitatively analyzing the "cooling marginal benefit" (i.e. the degree of influence of increasing or decreasing unit power on temperature drop) of each divided area under the current air conditioning power, the divided areas with low power utilization efficiency (stable areas) and high efficiency potential (non-stable areas) are identified. Through an iterative process, redundant power is recovered from stable areas and preferentially injected into non-stable areas with the highest marginal benefit, ultimately promoting the marginal benefit of all divided areas to approach, achieving the optimal state of the overall cooling effect of the system.

[0053] When the initial passengers have not yet been loaded, a unified air conditioning power can be set for each divided area according to historical data and external environmental temperature. As the train arrives at the station and the number of passengers increases, the air conditioning power of each divided area is adjusted in real time to meet the actual cooling demand.

[0054] In one embodiment, with reference to Figure 2 , step S3 includes steps S30-S32, as follows:

[0055] S30: For a single divided area, use the LSTM model to predict the predicted temperature drop time series when the air conditioning power is increased by a preset power and the predicted temperature drop time series when the air conditioning power is decreased by a preset power, calculate the sensitivity value of the corresponding divided area to the change in air conditioning power according to the two predicted temperature drop time series, and label the divided area as a stable area or a non-stable area according to the sensitivity value.

[0056] A small perturbation of increasing and decreasing a preset power is added to the current real-time air conditioning power of the divided area, and the LSTM model is used to output the predicted temperature drop time series after the air conditioning power is increased by the preset power and the predicted temperature drop time series after the air conditioning power is decreased by the preset power. In the predicted temperature drop time series corresponding to the air conditioning power increased by the preset power, a first temperature drop stable value is selected, and in the predicted temperature drop time series corresponding to the air conditioning power decreased by the preset power, a second temperature drop stable value is selected. Based on the first temperature drop stable value and the second temperature drop stable value, a central difference gradient is calculated to obtain the sensitivity value of the divided area. It can be seen that the central difference gradient is the same as the sensitivity value.

[0057] The calculation formula of the sensitivity value is as follows:

[0058]

[0059] In the formula, represents the central difference gradient of the i-th divided area; represents the central difference gradient of the i-th divided area; represents the central difference gradient of the i-th divided area; a first temperature drop stable value of the divided region; a second temperature drop stable value of the divided region; a second temperature drop stable value of the divided region; represents a preset power. The selection process of the first temperature drop stable value is as follows: in the corresponding predicted temperature drop time sequence, when the difference between the predicted temperature drop value at a certain time and the predicted temperature drop value at the previous adjacent time is less than 0.01 (which can be adjusted according to the situation), it is determined that the predicted temperature drop value at this time is the first temperature drop stable value. The selection process of the second temperature drop stable value is the same as that of the first temperature drop stable value, which will not be repeated.

[0060] The sensitive value reflects the change before and after the increase and decrease of power. If the sensitive value of the first divided region is greater than or equal to a preset fixed value (such as 0.1), it means that the region is very sensitive to power change, and increasing power can bring significant temperature drop improvement, which is a candidate area for power injection, and is marked as a non-stable region; if the sensitive value of the first divided region is less than the preset fixed value, it means that the refrigeration effect of the region has reached a saturation point, and the additional temperature drop benefit brought by increasing power is very low (even zero), so it is a candidate area for power recovery, and the used power may be excessive, which is marked as a stable region.

[0061] Similarly, all divided regions are marked as stable regions or non-stable regions.

[0062] S31: Set a to-be-allocated pool, for each stable region, extract a first power from the current air conditioner power to obtain a first adjusted power after weakening the first power, set an extraction condition according to the predicted temperature drop time sequence under the first adjusted power and the predicted temperature drop time sequence under the current air conditioner power, and in response to the extraction condition being met, put the first power into the to-be-allocated pool.

[0063] For stable regions, region power recovery is needed, and each time a first power is extracted from the current air conditioner power to obtain a first adjusted power after weakening the first power , the first power extracted each time is a fixed value. The smaller the value, the more the iteration times, but the more precise the final power obtained.

[0064] The extraction criteria are set as follows: An LSTM model is used to predict the predicted temperature drop time series under the first adjusted power and the current air conditioning power. A first stable value is selected from each of the two predicted temperature drop time series (the selection process is the same as for the first stable temperature drop value). The absolute value of the difference between the two first stable values ​​is calculated to obtain the temperature drop change value. When the temperature drop change value is less than a preset value, the extraction criteria are met. If the extraction criteria are met, the extracted first power is added to the allocation pool; otherwise, extraction is stopped.

[0065] The formula for calculating the temperature drop change is as follows:

[0066]

[0067] In the formula, Indicates the first Each region is divided into sections, and the temperature drop before and after the first power is extracted each time. Indicates the first The first stable value of the predicted temperature drop for each region under the first adjusted power condition; Indicates the first The first stable value of the predicted temperature drop for each divided region under the current air conditioning power conditions.

[0068] If the temperature drop change value Extremely small, such as the change in temperature drop. Less accurate than temperature sensor If the current division is stable, it can be considered stable, indicating that the current region is still in a stable state. Further power reduction is needed, so let the power of the first... Current air conditioning power in each zone (Power update), and extract the first power Placed into the allocation pool , (Power return), and a new round of iterations begins, until... The extraction stops, indicating that the minimum stable power has been obtained at this point. .

[0069] S32: For each unstable region, extract the second power from the allocation pool, add it to the current air conditioning power of the unstable region to obtain the second adjustment power, set update conditions according to the predicted temperature drop time series under the second adjustment power and the predicted temperature drop time series under the current air conditioning power, and update the current air conditioning power to the second adjustment power in response to the update conditions being met.

[0070] Since each unstable region has a different sensitivity value, before power allocation, the unstable regions are arranged in descending order of sensitivity value, and the air conditioning power of the unstable regions is adjusted in the order of arrangement, that is, the unstable regions with larger sensitivity values ​​are given priority for power increase.

[0071] Each time from the allocation pool Extracting the second power The current air conditioning power in the current unstable region With the second power The second adjustment power is obtained by adding them together. .

[0072] The update condition is set as follows: Using an LSTM model, predict the predicted temperature drop time series under the second adjusted power and the current air conditioning power. Select a second stable value from each of the two predicted temperature drop time series (the selection process is the same as for the first stable temperature drop value). Calculate the absolute value of the difference between the two second stable values ​​to obtain the temperature drop change value. When the temperature drop change value is greater than a preset value, the update condition is met. When the update condition is met, update the current air conditioning power to the second adjusted power; otherwise, stop updating.

[0073] The formula for calculating the temperature drop change is as follows:

[0074]

[0075] In the formula, Indicates the first The temperature drop changes before and after adding the second power to each divided region; Indicates the first The second stable value of the predicted temperature drop for each region under the second adjusted power condition; Indicates the first The second stable value of the predicted temperature drop for each region under the current air conditioning power conditions.

[0076] If the temperature drop change value This indicates that the power is not yet stable and needs to be increased further. Therefore, let the [missing information - likely a specific value or setting]... Current air conditioning power in each zone (Power Update), Pool to be Allocated (Power extraction), and a new round of iterations begins, until... At this point, a stable power is obtained. .

[0077] The second power increase each time Instead of a fixed value, before each power extraction from the allocation pool, the ratio of the current sensitivity value of the unstable region to the sum of the sensitivity values ​​of all partitioned regions is calculated. This ratio is then multiplied by the total power in the allocation pool to obtain the second power for this round. With the iteration process, the non-stable region gradually tends to be stable, that is, the sensitivity value is continuously reduced, and the ratio of the sensitivity value to the sum of the sensitivity values of all divided regions is also continuously reduced, and the total power of the power pool to be allocated is also continuously reduced, so the second power increased each time is also continuously reduced, thereby obtaining more accurate stable power .

[0078] The application adjusts the air conditioning power of each divided region dynamically, ensures that the high human flow density region obtains sufficient refrigeration power, and the low human flow density region avoids energy waste, and improves the overall resource utilization efficiency.

[0079] The application further provides an air conditioner intelligent operation and maintenance system for rail transit, which comprises a processor and a memory, and the memory stores computer program instructions.

[0080] The system further comprises a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.

[0081] It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the application, which are all within the protection scope of the application. Therefore, the protection scope of the application patent should be subject to the appended claims.

Claims

1. A rail transit-oriented air conditioning intelligent operation and maintenance method, characterized in that, The method comprises the following steps: Real-time acquisition of air conditioner power, passenger flow density and temperature in each divided area in the carriage; Construction of an LSTM model, wherein the input of the LSTM model is the time series of air conditioner power, passenger flow density and temperature, and the output is the time series of predicted temperature drop, and the air conditioner power of each divided area is adjusted, and the air conditioners of each divided area are operated at the adjusted power; The adjustment process comprises the following steps: for a single divided area, the LSTM model is used to predict the time series of predicted temperature drop when the air conditioner power is increased by a preset power and the time series of predicted temperature drop when the air conditioner power is reduced by the preset power, the sensitivity value of the corresponding divided area to the change of air conditioner power is calculated according to the two time series of predicted temperature drop, and the divided area is marked as a stable area or a non-stable area according to the sensitivity value; a to-be-allocated pool is set, in each stable area, a first power in the current air conditioner power is extracted to obtain a first adjusted power after the first power is weakened, an extraction condition is set according to the time series of predicted temperature drop under the first adjusted power and the time series of predicted temperature drop under the current air conditioner power, and in response to the extraction condition being met, the first power is put into the to-be-allocated pool; in each non-stable area, a second power in the to-be-allocated pool is extracted, and the second power is added to the current air conditioner power of the non-stable area to obtain a second adjusted power, an update condition is set according to the time series of predicted temperature drop under the second adjusted power and the time series of predicted temperature drop under the current air conditioner power, and in response to the update condition being met, the current air conditioner power is updated to the second adjusted power.

2. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 1, characterized in that, Before adjusting the air conditioner power of each divided area, the method further comprises the following steps: for a single divided area, the confidence of the output result of the LSTM model is calculated, and when the confidence is less than a preset correction threshold continuously for a set number of times, the dynamic parameters of the LSTM model are corrected, and when the number of corrections is greater than a preset number within a preset time, a maintenance work order is generated.

3. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 2, characterized in that, The confidence calculation process comprises the following steps: In the time series of predicted temperature drop output by the LSTM model, a predicted temperature drop value corresponding to the current time is selected, and the error between the predicted temperature drop value and the actual temperature drop value at the current time is calculated; a protection threshold is set, the ratio of the absolute value of the error to the maximum of the protection threshold and the predicted temperature drop value is calculated, and the difference between 1 and the ratio is taken as the confidence.

4. The rail transit-oriented air conditioning intelligent operation and maintenance method of claim 1, wherein The sensitivity value acquisition process comprises the following steps: a first temperature drop stability value is selected from the time series of predicted temperature drop when the air conditioner power is increased by the preset power, and a second temperature drop stability value is selected from the time series of predicted temperature drop when the air conditioner power is reduced by the preset power; the central difference gradient is calculated based on the first temperature drop stability value and the second temperature drop stability value to obtain the sensitivity value.

5. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 1, characterized in that, The sensitivity value is used to mark the divided area as a stable area or a non-stable area, which specifically comprises the following steps: when the sensitivity value is greater than or equal to a preset fixed value, the corresponding divided area is marked as a non-stable area; and when the sensitivity value is less than the preset fixed value, the corresponding divided area is marked as a stable area.

6. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 1, characterized in that, Before adjusting the air conditioner power of the non-stable area, the method further comprises the following steps: the non-stable areas are arranged in descending order according to the size of the sensitivity value, and the air conditioner power of the non-stable areas is adjusted in turn according to the arrangement order.

7. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 1, characterized in that, The setting method of the extraction condition is: selecting a first stable value from the predicted temperature drop time sequence under the first adjusted power and the predicted temperature drop time sequence under the current air conditioner power respectively, calculating the absolute value of the difference between the two first stable values to obtain a first temperature drop change value, and when the first temperature drop change value is less than a preset value, the extraction condition is met. 8.The rail transit-oriented air conditioner intelligent operation and maintenance method of claim 1, wherein, The setting method of the update condition is: selecting a second stable value from the predicted temperature drop time sequence under the second adjusted power and the predicted temperature drop time sequence under the current air conditioner power respectively, calculating the absolute value of the difference between the two second stable values to obtain a second temperature drop change value, and when the second temperature drop change value is greater than a preset value, the update condition is met.

9. The rail transit-oriented air conditioning intelligent operation and maintenance method according to claim 8, characterized in that, The obtaining process of the second power includes: calculating the ratio of the current sensitive value corresponding to the unstable region to the sum of the sensitive values of all divided regions, calculating the product of the ratio and the total power in the to-be-allocated pool to obtain the second power.

10. A rail transit-oriented air conditioning intelligent operation and maintenance system, characterized in that, The method comprises: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, the track-oriented air conditioner intelligent operation and maintenance method according to any one of claims 1-9 is realized.

Images