A tunnel intelligent ventilation control optimization method based on data fusion

Abstract

The application discloses a tunnel intelligent ventilation control optimization method based on data fusion, relates to the technical field of control optimization, and comprises the following steps: collecting tunnel environment data of multiple sensing devices in a tunnel, and performing anti-noise sparse coding processing on the tunnel environment data to obtain a stable fusion environment sensing data set; loading the stable fusion environment sensing data set into a preset fusion model pool to perform fusion reasoning and state calculation, and obtaining an environment state fusion data set; dynamically adjusting the weight distribution of each fusion model in the fusion model pool based on the environment state fusion data set, and obtaining an environment state evaluation result set; adjusting the fusion model weight and parameters through real-time control feedback based on sensing device operation feedback data and in combination with the environment state evaluation result set, and generating a standardized control execution instruction set; and through the synergistic effect, the ventilation control system continuously outputs accurate environment state evaluation results and standardized control instructions, and dynamic optimization regulation and control of ventilation parameters are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control optimization, in particular to a tunnel intelligent ventilation control optimization method based on data fusion. BACKGROUND

[0002] As a core link of safe operation and maintenance of transportation infrastructure, tunnel intelligent ventilation control technology mainly collects tunnel internal environment parameters through deploying multi-source environment perception devices (such as carbon monoxide concentration sensors, temperature and humidity monitors, wind speed and direction instruments, etc.), and estimates the environment state based on a preset physical model or statistical fusion algorithm (such as Kalman filtering, weighted average fusion, etc.), and finally generates ventilation device control instructions. The current technical route generally adopts a fixed weight data fusion mechanism, that is, the fusion coefficients of each perception parameter are determined through historical data training, or a single environment evaluation model is used to output the control strategy. Such methods can effectively maintain the pollutant concentration and temperature and humidity levels in the tunnel under steady-state conditions, and meet the basic requirements of the industry standard for the response speed and energy consumption control of the ventilation system.

[0003] However, the existing methods have limitations in dealing with dynamic running environments of tunnels: on the one hand, the fixed weight fusion mechanism is difficult to adapt to nonlinear disturbances such as traffic flow mutations and vehicle exhaust emissions fluctuations, resulting in that the environment evaluation results lag behind the actual working condition changes; on the other hand, there is a lack of real-time optimization capability for multi-model collaborative mechanism, when a single model produces estimation deviation due to sensor drift or local environmental abnormalities, the overall evaluation accuracy cannot be maintained through weight redistribution and parameter online calibration, affecting the robustness of ventilation control. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a tunnel intelligent ventilation control optimization method based on data fusion to solve the problems of poor adaptability and insufficient model collaborative optimization of existing methods in dynamic environments.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a tunnel intelligent ventilation control optimization method based on data fusion, which comprises,

[0008] Collecting tunnel environment data of multiple perception devices in the tunnel, and performing anti-noise sparse coding processing on the tunnel environment data to obtain a stable fusion environment perception data set;

[0009] Loading the stable fusion environment perception data set into a preset fusion model pool for fusion reasoning and state calculation to obtain an environment state fusion data set;

[0010] The weight allocation of each fusion model in the fusion model pool is dynamically adjusted based on the environmental state fusion data set to obtain the environmental state assessment result set.

[0011] Based on the operational feedback data of sensing devices and combined with the set of environmental status assessment results, the weights and parameters of the fusion model are adjusted through real-time control feedback to generate a standardized set of control execution instructions.

[0012] The ventilation operation parameters of multiple sensing devices in the tunnel are adjusted based on a standardized set of control execution instructions to generate optimized ventilation operation parameters.

[0013] As a preferred embodiment of the tunnel intelligent ventilation control optimization method based on data fusion described in this invention, the steps of collecting tunnel environmental data from multiple sensing devices within the tunnel and performing noise-resistant sparse coding processing on the tunnel environmental data to obtain a stable fused environmental sensing data set are as follows.

[0014] Collect multidimensional raw environmental parameter data, and perform time alignment processing on the multidimensional raw environmental parameter data to generate a set of raw multidimensional tunnel environmental parameter data after time alignment;

[0015] The original multidimensional tunnel environment parameter data set after time alignment is divided according to the set time window length to generate multiple environment data processing windows containing complete and continuous time series data.

[0016] Based on the environmental data processing window, a time series feature vector is constructed from the original multidimensional tunnel environmental parameter data set, and a sparse coding method is used to perform sparse modeling in the preset feature base representation space to generate a multidimensional tunnel environmental parameter data set after sparse representation reconstruction.

[0017] The difference between the multidimensional tunnel environment parameter data set reconstructed by sparse representation and the original multidimensional tunnel environment parameter data is calculated. Data with a difference amplitude exceeding the preset range are identified as distorted data segments and removed. Then, the sparse representation data after removing distorted points is normalized to generate a stable fused environmental perception data set.

[0018] As a preferred embodiment of the intelligent ventilation control optimization method for tunnels based on data fusion described in this invention, the multidimensional original environmental parameter data includes gas concentration parameters, aerodynamic parameters, thermal temperature environmental parameters, traffic flow parameters, and background environmental parameters.

[0019] As a preferred embodiment of the tunnel intelligent ventilation control optimization method based on data fusion described in this invention, the steps of substituting a stable fused environmental perception data set into a preset fusion model pool for estimation and fusion calculation to obtain an environmental state fused data set are as follows:

[0020] The stable fusion environment perception data set is substituted into the preset fusion model pool, multi-dimensional environment state independent estimation operation is performed, and an independently estimated environment state data set is generated;

[0021] The independently estimated environment state data set is structured and collected to construct a structured basic environment state estimation data set;

[0022] According to the preset fusion strategy, the structured basic environment state estimation data set is multi-dimensionally weighted and integrated to generate a final environment state fusion data set for feedback regulation.

[0023] As a preferred scheme of the tunnel intelligent ventilation control optimization method based on data fusion, wherein: the preset fusion model pool refers to an independent mathematical model set constructed according to the environmental parameters and actual working conditions of the tunnel environment.

[0024] As a preferred scheme of the tunnel intelligent ventilation control optimization method based on data fusion, wherein: the weight distribution of each fusion model in the fusion model pool is dynamically adjusted based on the environment state fusion data set, and an environment state evaluation result set is obtained, and the steps are as follows,

[0025] The typical response interval reflecting the matching relationship between the environment state and the control instruction is extracted as the performance evaluation benchmark of the fusion model by using the tunnel operation historical data and the artificial auditing and labeling results.

[0026] Based on each fusion model in the fusion model pool, the error deviation between the fusion model evaluation value and the standard state reference value is calculated, and the error deviation is mapped to the relative confidence value of each fusion model.

[0027] Based on the updating strategy of the exponentially weighted moving average, the relative confidence value of each fusion model is dynamically adjusted, and all the relative confidence values are normalized to generate a dynamic participation weight set of each fusion model under the current working condition.

[0028] Based on the dynamic participation weight set of each fusion model under the current working condition, the environment state estimation results of each fusion model in the fusion model pool under the current working condition are weighted and summarized to generate an environment state evaluation result set under the cooperation of multiple fusion models.

[0029] As a preferred scheme of the tunnel intelligent ventilation control optimization method based on data fusion, wherein: based on the perception device operation feedback data, the environment state evaluation result set is combined to adjust the fusion model weight and parameter through real-time control feedback, and a standardized control execution instruction set is generated, and the steps are as follows,

[0030] The perception device operation feedback data is time-synchronized and matched with the environment state evaluation result set, a feedback data set based on the perception device state change and the environment state response condition is constructed;

[0031] According to the feedback data set based on the perception device state change and the environment state response condition, the adaptation ability of each fusion model in the fusion model pool is judged, and the participation weight of each fusion model is dynamically adjusted according to the adaptation ability, and a fusion model participation weight set after weight dynamic update is obtained;

[0032] According to the error performance in the fusion model participation weight set after weight dynamic update, the internal parameters of each fusion model in the fusion model pool are directionally fine-tuned, and a fusion model set after weight adjustment and parameter optimization is obtained;

[0033] Based on the fusion model set after weight adjustment and parameter optimization, the current environment state is re-evaluated, and a standardized control execution instruction set is constructed.

[0034] As a preferred scheme of the tunnel intelligent ventilation control optimization method based on data fusion, wherein: according to the standardized control execution instruction set, the ventilation operation parameters of the plurality of perception devices in the tunnel are adjusted, and an optimized ventilation operation parameter set is generated, the steps are as follows,

[0035] According to the target state value interval in the standardized control execution instruction set, the operation parameter type that needs to be adjusted for each ventilation device is extracted and mapped, and a corresponding relationship between the device parameter and the target state interval is constructed as a device parameter target state mapping table;

[0036] According to the device parameter target state mapping table, combined with the operation parameter values currently collected by each ventilation device, the deviation values between each parameter and the target state interval are calculated, and a device parameter adjustment reference deviation set is generated;

[0037] According to the device parameter adjustment reference deviation set, a control adjustment strategy is adopted to determine the specific adjustment direction and adjustment amplitude of each parameter, and a target adjustment action sequence of the device operation parameter is generated;

[0038] According to the target adjustment action sequence of the device operation parameter, adjustment instructions are issued to the corresponding ventilation device control interface, and the device response state is collected, and an optimized ventilation operation parameter set is generated.

[0039] The application has the beneficial effects that: through the step of real-time control feedback adjustment fusion model weight and parameter based on the sensing device operation feedback data, the model weight and internal parameters are fine-tuned according to the device state feedback data, a closed-loop optimization mechanism is established, and the fault tolerance and online learning ability of the model to local anomalies are enhanced. The synergistic effect enables the ventilation control system to continuously output accurate environmental state evaluation results and standardized control instructions, and realizes dynamic optimization and regulation of ventilation parameters. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0041] Fig. 1 Flow chart of the tunnel intelligent ventilation control optimization method based on data fusion.

[0042] Fig. 2 Flow chart of the tunnel intelligent ventilation control optimization method.

[0043] Fig. 3 Flow chart of the dynamic weight adjustment of the fusion model pool.

[0044] Fig. 4 Flow chart of the real-time control feedback adjustment. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0046] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0047] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0048] REFERENCE Figs. 1-4 For one embodiment of the present application, the embodiment provides a tunnel intelligent ventilation control optimization method based on data fusion, comprising the following steps:

[0049] S1: Collect tunnel environment data of multiple sensing devices in the tunnel, and perform anti-noise sparse coding processing on the tunnel environment data to obtain a stable fused environment perception data set.

[0050] Specifically, the steps are as follows,

[0051] S1.1: Multiple environment sensing devices are arranged in different areas in the tunnel, and multi-dimensional original environment parameter data is synchronously acquired according to a preset data sampling frequency, and time alignment processing is performed on the multi-dimensional original environment parameter data to generate a set of original multi-dimensional tunnel environment parameter data after time alignment.

[0052] Specifically, multiple environment sensing devices are arranged at multiple position areas in the tunnel, and by setting a unified sampling frequency parameter, the multi-dimensional environment parameters of each environment sensing device are synchronously collected, and after collection, all collected multi-dimensional tunnel environment parameter data is aligned according to a unified time index. Linear interpolation method is used to process data points with differences between data timestamps, so that the multi-dimensional tunnel environment parameter data collected by different environment sensing devices have a strict one-to-one correspondence in the time axis. Finally, a set of original multi-dimensional tunnel environment parameter data after time alignment is generated.

[0053] Among them, the multi-dimensional environment parameters specifically include but are not limited to: carbon monoxide concentration, nitrogen oxide concentration, air temperature, air humidity, and wind speed and direction, etc.

[0054] It needs to be explained that the unified sampling frequency parameter is selected according to the tunnel operating environment change rate and the upper limit of the data collection frequency of the environment sensing device, and a fixed sampling interval value is selected by a preset rule; The unified time index is to take the starting time point of collection as the starting point of the time index sequence, and a set of standard time stamp set is formed by equidistant recursion according to the fixed sampling interval after determining the sampling frequency parameter.

[0055] It needs to be explained that the linear interpolation method is used when there is a deviation between the actual timestamp of the multi-dimensional tunnel environment parameter data in the collection process and the unified time index. Linear interpolation method is based on the linear relationship between the two adjacent multi-dimensional tunnel environment parameter data with clear timestamp. It is assumed that the tunnel environment parameters of each dimension change linearly with time between adjacent time points. And by using the linear interpolation method, the multi-dimensional tunnel environment parameter data collected by all sensing devices can be uniformly time-aligned, ensuring that the set of original multi-dimensional tunnel environment parameter data after time alignment has time consistency.

[0056] S1.2: The time-aligned original multi-dimensional tunnel environment parameter data set is divided according to the set time window length, and a plurality of environment data processing windows containing complete continuous time sequence data are generated.

[0057] Specifically, the time-aligned original multi-dimensional tunnel environment parameter data set is segmented according to the preset time window length, and a sliding window method is used to step from the start time index, and data sequences with the same length as the time window length are sequentially intercepted, and it is ensured that each window contains continuous and complete time index sequences and corresponding multi-dimensional tunnel environment parameter data contents. The multi-dimensional tunnel environment parameter data contents are used as independent environment data processing windows for time series analysis and feature extraction operations, and finally a plurality of environment data processing windows containing complete continuous time sequence data are generated.

[0058] It needs to be explained that the preset time window length refers to the fixed time interval length set for intercepting each continuous time sequence data when dividing the time-aligned original multi-dimensional tunnel environment parameter data set. The preset time window length is based on the matching relationship between the periodicity analysis result of the tunnel environment state change and the sampling frequency of the perception device. After comprehensive judgment on the fluctuation frequency analysis, event response delay evaluation, and ventilation control action cycle identification of historical multi-dimensional tunnel environment parameter data, the time interval that can completely cover the typical environment change process is selected as the time window length.

[0059] It needs to be explained that time series analysis refers to the process of sequence structure identification and change trend modeling of multi-dimensional tunnel environment parameter data in each environment data processing window with time index as the sequence. It usually includes trend judgment, periodicity detection, and change rate operations, which are used to identify the dynamic evolution characteristics of the environment state. Feature extraction operation refers to extracting representative numerical features from multi-dimensional tunnel environment parameter data based on time series analysis.

[0060] S1.3: According to the environment data processing window, the tunnel environment parameter change sequence in the time dimension is extracted from the original multi-dimensional tunnel environment parameter data set, and a time series feature vector is constructed. Then, based on the time series feature vector, a sparse coding method is used to perform sparse modeling in the preset feature basis representation space, and a sparse reconstruction multi-dimensional tunnel environment parameter data set is obtained.

[0061] Specifically, according to each environment data processing window, the variation sequence of each dimension tunnel environment parameter dimension in the time dimension is extracted from the original multi-dimensional tunnel environment parameter data set aligned in time from the completion time in time index order, and each environment data processing window corresponding time sequence feature vector is combined and constructed, then the time sequence feature vector is taken as the input to construct the sparse coding model, and based on the preset feature basis representation space, the sparse representation coefficient of the time sequence feature vector is obtained by solving the sparse optimization process of minimizing the reconstruction error, and finally the sparse representation reconstruction is completed by using the sparse representation coefficient and the preset feature basis representation space to reconstruct, and finally the multi-dimensional tunnel environment parameter data set after sparse expression reconstruction is obtained.

[0062] It needs to be explained that the sparse coding model is to take the time sequence feature vector as the training input, construct the objective function of minimizing the reconstruction error, and combine the set feature basis dimension and sparse expression constraint to construct the sparse representation structure of the tunnel environment state. In the process of sparse coding model training, based on the multi-dimensional tunnel environment parameter data aligned in completion time, the time sequence feature vector sample set is constructed in time order as the training sample data set, and the sparse representation coefficient of the training sample data set is calculated under the premise of keeping the feature basis dimension unchanged, then the feature basis is optimized in reverse to further reduce the reconstruction error, the sparse coding model training is completed, and finally the data transformation ability for sparse expression modeling and reconstruction is obtained.

[0063] It needs to be explained that the preset feature basis representation space refers to a set of representative data variation benchmark sequences in the time dimension before the sparse coding model is constructed, which is determined by statistical analysis according to the variation trend and typical fluctuation mode of a large amount of historical tunnel environment parameter data in the time dimension.

[0064] S1.4: According to the multi-dimensional tunnel environment parameter data set after sparse expression reconstruction, the difference amplitude calculation is performed with the original multi-dimensional tunnel environment parameter data, and the original multi-dimensional tunnel environment parameter data with a difference amplitude exceeding the preset range is determined as a distorted point data segment and is removed, then the remaining original multi-dimensional tunnel environment parameter data is normalized, and finally a stable fusion environment perception data set is generated.

[0065] Specifically, based on the sparse expression reconstructed multi-dimensional tunnel environment parameter data set, the original multi-dimensional tunnel environment parameter data on the time axis corresponding to the sparse expression reconstructed multi-dimensional tunnel environment parameter data is aligned point by point, the absolute difference between the corresponding values is calculated item by item according to the same parameter item and time index, a difference amplitude sequence is constructed, then based on the preset difference amplitude tolerance range, the difference amplitude sequence is discriminated, when the difference amplitude of any dimension in a certain time period exceeds the preset difference amplitude tolerance range, the original multi-dimensional tunnel environment parameter data in the corresponding time period is marked as a distorted point data segment and is removed, after completing the removal of the distorted point data segment, the remaining original multi-dimensional tunnel environment parameter data is interval normalized according to the historical statistical upper and lower bounds of each dimension, to generate fusion data with stable numerical distribution and consistent physical quantity scale, and finally to obtain a stable fusion environment perception data set.

[0066] It needs to be explained that the preset difference amplitude tolerance range is set based on the historical fluctuation characteristics of the tunnel environment parameters and the acceptable variation range of the physical quantity. In the setting process, by statistical analysis of the historical data, the fluctuation amplitude of different environment parameters under normal operating conditions is evaluated, and a reasonable tolerance range is determined.

[0067] S2: Substitute the stable fusion environment perception data set into the preset fusion model pool for fusion reasoning and state calculation to obtain an environment state fusion data set.

[0068] Specifically, the steps are as follows,

[0069] S2.1: Substitute the stable fusion environment perception data set into the preset fusion model pool, perform multi-dimensional environment state independent estimation operation, and generate an independent estimated environment state data set.

[0070] Specifically, substitute the stable fusion environment perception data set into the preset fusion model pool, based on each stable fusion environment perception data in the stable fusion environment perception data set, respectively use the time series deduction method and multi-dimensional environment correlation estimation method embedded in the fusion model to extract the dynamic change trend and future estimated value of each environment parameter in a short time, and take the future estimated value as the independent estimation result of the current time, then based on the mapping relationship between each environment parameter and the related fusion model corresponding to the environment parameter, use the environment coupling influence causal analysis method to extract key influence features from the estimated results, such as short sequence trend slope, synergistic offset between environment parameters and environment disturbance response features, etc., and combine the preset state mapping table to estimate the independent estimation value of each environment parameter, to generate an independent estimated environment state data set.

[0071] It needs to be explained that the preset fusion model pool refers to a set of independent mathematical model collection constructed according to various environmental parameters of the tunnel environment (such as carbon monoxide concentration, nitrogen oxide concentration, temperature and humidity, wind speed and direction, etc.) and actual working conditions (such as traffic flow, equipment running state, etc.). In the construction stage of each fusion model in the fusion model pool, according to the physical characteristics of different environmental parameters and the coupling relationship of working conditions (for example: the diffusion of carbon monoxide concentration, the time sequence dependence of temperature and humidity, and the fluid dynamics characteristics of wind speed and direction), the corresponding model is constructed through algorithm mapping rules; The training process of the preset fusion model pool is based on a large amount of historical tunnel environment data, experimental data and actual working conditions. Each fusion model is trained for a specific type of environmental parameter (such as carbon monoxide concentration, nitrogen oxide concentration, air temperature, etc.). The training process includes statistical analysis of historical data, application of machine learning algorithms (such as regression analysis, deep learning, Kalman filtering, etc.), and supervised learning through evaluation reference results confirmed by artificial review. In the training stage of each fusion model in the fusion model pool, the steady-state historical data is used to initialize the basic parameters of each model, and the mean absolute error is used as the loss function for physical parameters, the initial error index is minimized, the dynamic time warping distance is optimized for time sequence parameters to accurately fit the change trend, and simulated disturbance data is injected. Through the adversarial training mechanism, the anti-interference ability of the model is improved, the model remains stable in the dynamic environment, and the evaluation results meet the actual safety standards.

[0072] It needs to be explained that the evaluation reference results confirmed by artificial review refer to high-credibility evaluation labels used for training supervised models through in-depth analysis of tunnel historical environmental data and ventilation response effects by operation and maintenance personnel. The evaluation reference results confirmed by artificial review include the optimal response parameters of ventilation control under a certain environmental state, the ventilation equipment running set value under the known safety boundary condition, or the effective ventilation regulation strategy verified in the accident plan drill, etc.

[0073] It needs to be explained that the mapping relationship between the relevant fusion models corresponding to the environmental parameters refers to the mapping rules and functional connections between each environmental parameter (such as carbon monoxide concentration, air temperature, humidity, etc.) and a specific model in the preset fusion model pool, which includes: matching of environmental parameters and models, fusion model functions, and estimation process of independent estimated values of each environmental parameter.

[0074] It needs to be explained that the preset state mapping table is to perform clustering analysis on the environmental perception data in the stable fusion environmental perception data set, extract representative environmental state sample categories, and statistically model the parameter value intervals corresponding to the environmental state sample categories. Then, combined with domain knowledge and environmental safety management specifications, the grade label corresponding to each environmental state sample category is obtained, and each class of grade label and the corresponding parameter value interval is archived, and finally the mapping relationship between the parameter value interval and the environmental state grade is generated as the state mapping table.

[0075] Among them, the domain knowledge and environmental safety management specifications mainly refer to the existing engineering practical experience, environmental response law summary, sensor application standard and control strategy criterion and other technical knowledge system in the field of tunnel ventilation and environmental control, as well as the relevant standards and management specifications of tunnel operation safety, environmental quality control, ventilation facility operation requirements and other relevant standards and management specifications issued by the state or industry.

[0076] S2.2: Structured collection of independently estimated environmental state data set to construct structured basic environmental state estimation data set.

[0077] Specifically, based on the independently estimated environmental state data set, the time index and the environmental parameter estimation value corresponding to the time index are extracted from each independently estimated environmental state data, and the environmental parameter estimation value corresponding to the time index is unified and integrated according to time sequence, and a structured basic environmental state estimation data set is constructed.

[0078] For example, assuming that the stable fusion environmental perception data set contains temperature, humidity, gas concentration and other environmental parameters, the estimation values of temperature, humidity and gas concentration in the independently estimated environmental state data set will be integrated into a data record at the corresponding time point. After unified formatting, the environmental parameter estimation value at each time point is paired according to the time index to ensure that the units and representations of all data are consistent, such as temperature (℃), humidity (%), gas concentration (ppm), etc., which are converted into a standard structured basic environmental state estimation data set.

[0079] S2.3: According to the preset fusion strategy, the structured basic environmental state estimation data set is integrated in multiple dimensions, and the final environmental state fusion data set for feedback control is generated.

[0080] Specifically, the environmental parameter estimation values of each time dimension are extracted from the structured basic environmental state estimation data set, and different weights are given to each environmental parameter estimation value according to the preset fusion strategy, and the weighted environmental parameter estimation values are obtained, and then the weighted environmental parameter estimation values are combined according to the time sequence, and the final environmental state fusion data set for feedback control is generated.

[0081] It needs to be explained that the preset fusion strategy refers to a set of fusion rules for estimating values of various environmental parameter estimation data in the structured basic environment state estimation data set according to the regulation and control requirements of the tunnel environment control. The preset fusion strategy includes: environmental parameter weight distribution rule, time sequence fusion method and estimation value abnormality suppression mechanism.

[0082] S3: dynamically adjusting the weight distribution of each fusion model in the fusion model pool based on the environment state fusion data set, and obtaining an environment state evaluation result set.

[0083] Specifically, the steps are as follows,

[0084] S3.1: using tunnel operation history data and artificial audit evaluation annotation results, extracting a typical response interval reflecting the matching relationship between the environment state and the control instruction as the performance evaluation benchmark of the fusion model.

[0085] Specifically, based on the tunnel operation history data and the artificial audit evaluation annotation results, the environmental parameter record sequence and the ventilation control execution record sequence in the tunnel operation history data are extracted, and the relationship between the typical tunnel environment state and the ventilation control effect is explained in combination with the artificial audit annotation results, to construct a tunnel operation state evaluation sample set. In the tunnel operation state evaluation sample set, based on the matching relationship between different environment state conditions and the corresponding ventilation control effect, an evaluation index space composed of multi-dimensional environment state indicators and control feedback effect indicators is constructed, and the numerical distribution characteristics of each dimension index in the evaluation index space in the tunnel operation state evaluation sample set are counted. The mean value performance of each tunnel operation state evaluation sample in the index interval of the tunnel operation state evaluation sample set and the dispersion degree between adjacent samples are obtained as the response consistency and representativeness between the environment state and the control feedback. The typical running state samples with representativeness, stability and response accuracy are screened out, and finally the matching relationship between the environmental parameter value range and the ventilation control parameter contained in the typical running state samples is extracted from the typical running state samples with representativeness, stability and response accuracy. The typical response interval reflecting the environment state and the control instruction in the matching relationship between the environmental parameter value range and the ventilation control parameter is taken as the performance evaluation benchmark of the fusion model.

[0086] The formula of the interval mean is as follows,

[0087]

[0088] Wherein, c represents the index of the environment state category, d represents the dimension index of the environmental parameter, i represents the index of the tunnel operation state evaluation sample, represents the interval mean of the dth environmental parameter dimension in the cth environment state category, N crepresents the total number of tunnel operation state evaluation samples in the cth environmental state category, represents the environmental parameter value of the ith tunnel operation state evaluation sample in the cth environmental state category in the dth environmental parameter dimension.

[0089] It needs to be explained that the relationship between the typical tunnel environmental state and the ventilation control effect means that in the manual audit annotation process, according to the environmental parameter record sequence and the ventilation control execution record sequence in the tunnel operation history data, combined with the annotation and evaluation results in the manual audit annotation results, the comprehensive judgment and description of the ventilation effect (such as the reduction of pollutant diffusion rate, the control of energy consumption, and the maintenance of personnel comfort, etc.) generated by the control response (such as fan start-stop state, speed gear, and air exchange frequency, etc.) of the ventilation system under specific environmental parameters (such as carbon monoxide concentration range, wind speed level, and traffic flow density, etc.) are made.

[0090] It needs to be explained that the matching relationship between different environmental state conditions and corresponding ventilation control effects means that in the tunnel operation process, each specific environmental parameter (for example: carbon monoxide concentration, nitrogen oxide concentration, air visibility, wind speed and direction, etc. State combination composed of multiple environmental parameters) will correspond to one or a group of ventilation control strategies (for example: fan start-stop state, fan running gear, air volume output, etc.), and the environmental improvement effect (such as pollutant concentration reduction amplitude, recovery time, etc.) generated by the ventilation control strategy in actual operation is verified by manual audit annotation or historical operation record, as an effective or representative typical ventilation control response effect.

[0091] It needs to be explained that the matching relationship between the environmental parameter value range and the ventilation control parameter is that in the tunnel operation process, when the environmental parameters (such as carbon monoxide concentration, nitrogen oxide concentration, air visibility, and wind speed and direction, etc.) are in a specific numerical range combination, the ventilation control parameter set that can effectively reduce the pollutant concentration or achieve the control of pollutant recovery time is verified by manual audit annotation and historical operation data; wherein, the specific numerical range combination refers to the specific environmental parameter interval determined by data statistics and manual audit experience summary for each environmental parameter in the tunnel in actual operation, which includes but is not limited to: carbon monoxide concentration interval, nitrogen oxide concentration interval, space visibility interval, and wind speed and direction interval.

[0092] It needs to be explained that the interval mean method refers to clustering and classifying the tunnel operation state evaluation samples in multiple environmental parameter dimensions in the tunnel operation state evaluation sample set. According to a preset environmental parameter value division method, the value distribution of each environmental parameter is divided into several continuous value intervals, and the mean value of the environmental parameter value in each value interval is calculated. The obtained environmental parameter interval mean value is used as the representative value of the environmental state feature, which is used to measure the concentration and representativeness of the environmental state sample in different value intervals; the outlier discrimination method refers to identifying the tunnel operation state evaluation sample records that significantly deviate from the main distribution trend in a certain environmental parameter dimension or multiple environmental parameter dimensions through multidimensional statistical analysis of each environmental parameter dimension in the tunnel operation state evaluation sample set, and determining the tunnel operation state evaluation sample with abnormal ventilation control response effect or unstable tunnel operation state.

[0093] S3.2: Based on each fusion model in the fusion model pool, calculate the error deviation between the fusion model evaluation value and the standard state reference value, and map the error deviation to the relative confidence value of each fusion model.

[0094] Specifically, the environmental state data at the current time point is input into each fusion model in the fusion model pool, the corresponding fusion model evaluation value is obtained, and the error deviation between the fusion model evaluation value and the standard state reference value is calculated to obtain the error deviation result. Then, according to the preset error mapping function, the error deviation is converted into the relative confidence value of each fusion model.

[0095] The conversion formula of the relative confidence value is as follows,

[0096] γ k = ψ (|f k (X t )- ρ|) ;

[0097] Wherein, k represents the index of the fusion model evaluation value, ρ represents the standard state reference value of the target monitored physical quantity, γ k represents the relative confidence value of the kth fusion model in the fusion model pool, t represents the time index of the current time point, X t represents the environmental state data collected at time t, f k (·) represents the fusion model evaluation value of the kth fusion model, |∈ k - ρ| represents the error deviation result of the same target monitored physical quantity, and ψ(·) represents the error mapping function, which maps the error deviation to the relative confidence.

[0098] It needs to be explained that the preset error mapping function is usually constructed according to the statistical relationship between error deviation and fusion model confidence level in historical monitoring data. The evaluation error of the fusion model under multiple typical scenes is normalized in the training stage, and is fitted with the confidence level evaluated by artificial labeling or expert experience, to form a monotonically decreasing mapping relationship.

[0099] S3.3: An updating strategy based on exponential weighted moving average, which dynamically adjusts the participation weight in the relative confidence value of each fusion model, and integrates to generate a dynamic participation weight set of each fusion model under the current working condition.

[0100] Specifically, based on the updating strategy of exponential weighted moving average, the relative confidence value of each fusion model at the current time is obtained, and the historical participation weight value of each fusion model is iteratively updated by using the exponential weighted moving average formula, the participation weight value of each fusion model at the current time is calculated, then the current participation weight value of the fusion model is normalized according to the participation weight value, and finally a dynamic participation weight set of each fusion model under the current working condition is integrated and generated.

[0101] The exponential weighted moving average formula is as follows,

[0102] θ q =λ·ζ q +(1-λ)·θ q-1 ;

[0103] Wherein, q represents the update round index of the current fusion model weight, θ q represents the participation weight value obtained after the qth update, λ represents the smoothing coefficient of exponential weighting, and needs to satisfy λ∈[0,1], ζ q represents the relative confidence value estimated and obtained in the qth update, θ q-1 represents the participation weight value reserved after the (q-1)th update.

[0104] It needs to be explained that the updating strategy of exponential weighted moving average is a dynamic adjustment method by assigning different weight proportions to the historical participation weight value of the fusion model and the relative confidence value of the fusion model at the current time, and performing weighted calculation.

[0105] S3.4: Based on the dynamic participation weight set of each fusion model under the current working condition, the environmental state estimation results of each fusion model in the fusion model pool under the current working condition are weighted and summarized to generate an environmental state evaluation result set under the cooperation of multiple fusion models.

[0106] Specifically, the environmental state estimation results and dynamic participation weights corresponding to each fusion model in the fusion model pool are extracted, the influence proportion of the environmental state estimation results of each fusion model is adjusted according to the dynamic participation weight, then the adjusted environmental state estimation results of all fusion models are combined to generate an environmental state evaluation result set under the cooperation of multiple fusion models.

[0107] For example, at a certain running time, multiple fusion models in the fusion model pool respectively perform state estimation on key environmental parameters such as temperature, humidity, carbon monoxide concentration and carbon dioxide concentration in the tunnel, and generate corresponding estimation results. At the same time, each fusion model has a specific dynamic participation weight based on the current working condition. In addition, according to the respective dynamic participation weight, the influence proportion of the estimation results of temperature, humidity and gas concentration output by each model is adjusted, and then the adjusted estimation results are structurally combined to form an environmental state evaluation result set under the cooperation of multiple models under the current working condition.

[0108] S4: Based on the perception device operation feedback data, combining the environmental state evaluation result set, adjusting the fusion model weight and parameter through real-time control feedback to generate a standardized control execution instruction set.

[0109] Specifically, the steps are as follows,

[0110] S4.1: Time synchronization matching is performed between the perception device operation feedback data and the environmental state evaluation result set to construct a feedback data set based on the state change of the perception device and the response of the environmental state.

[0111] Specifically, the time stamp information corresponding to each group of data records in the perception device operation feedback data is extracted, and the evaluation time stamp information corresponding to each environmental state estimation result in the environmental state evaluation result set is extracted. Under a unified time coordinate axis, the perception device operation feedback data and the environmental state evaluation result set are paired according to the time stamp consistency principle. The environmental state estimation result with the smallest time difference is selected as the matching item corresponding to the current perception device operation feedback data. After completing the pair-by-pair matching, the state change information field of the perception device and the corresponding environmental state response information field are extracted and data splicing is performed to generate a feedback information data with a unified structure. Finally, the time sequence order is arranged to construct a standardized control execution instruction set.

[0112] It needs to be explained that the timestamp precision consistency principle refers to ensuring that the time stamps of the data records of the two parties use the same time unit and precision (for example, both use pure digital continuous form to represent) when performing time synchronization matching between the perception device running feedback data and the environmental state evaluation result set, so as to avoid inaccurate comparison of time difference or incorrect pairing due to inconsistent time format or precision.

[0113] S4.2: According to the feedback data set based on the perception device state change and the environmental state response, the adaptation ability of each fusion model in the fusion model pool is judged, and the participation weight of each fusion model is dynamically adjusted according to the adaptation ability, and a fusion model participation weight set that has completed weight dynamic update is obtained.

[0114] Specifically, the perception device state change value and the corresponding environmental state response value in each group of information pairs are extracted from the feedback data set, and according to the state response prediction mechanism of each fusion model in the fusion model pool, the environmental state index value under the current environmental condition is predicted according to the perception device state change value, and all environmental state index values are summarized to generate a target environmental state index value set. Then, the target environmental state index value set is compared with the actual environmental state response value to obtain the adaptation degree index value of each fusion model. Finally, according to the preset weight dynamic update criterion, the participation weight of each model in the fusion model pool is adjusted to build a fusion model participation weight set that has completed weight dynamic update.

[0115] It needs to be explained that the state response prediction mechanism refers to the functional mapping relationship inside the fusion model for deducing the corresponding environmental state index value under the condition of inputting the perception device state change value. The state response prediction mechanism is usually obtained by supervised training on the perception device state change and environmental response matching sample pairs in the historical data during the establishment of the fusion model.

[0116] It needs to be explained that the preset weight dynamic update criterion determines the increase or decrease of the participation weight of the fusion model according to the prediction accuracy of the fusion model under the current environmental state. The preset weight dynamic update criterion can be realized by setting threshold, weighting coefficient or proportion rule, so as to ensure that the model with higher adaptation ability obtains greater participation weight, while the weight of the model with lower adaptation ability is reduced, so that the weight distribution in the fusion model pool is more reasonable and dynamic. The preset weight dynamic update criterion includes the following key elements: adaptation degree index, threshold setting, weighting coefficient, proportion rule and smoothing adjustment.

[0117] S4.3: According to the error performance reflected in the fusion model participation weight set that has completed weight dynamic update, the internal parameters of each fusion model in the fusion model pool are directionally fine-tuned to obtain a fusion model set that has completed weight adjustment and parameter optimization.

[0118] Specifically, the weight update information of each fusion model is extracted from the set of fusion models that have completed dynamic weight updating, the fusion model with more obvious error performance is identified, and the internal parameters of each fusion model are adjusted according to the error performance, so that each fusion model can better adapt to the environmental state evaluation requirements under the current working condition, and finally the set of fusion models after weight adjustment and parameter optimization is obtained.

[0119] It should be explained that the weight update information of each fusion model refers to recording the participation weight adjustment history of each fusion model in the process of dynamic weight updating of the fusion model, and the weight update information includes the weight value change at the current time, the weight change amplitude, the update direction and the frequency.

[0120] It should be explained that the internal parameters of each fusion model refer to the parameter set inside each fusion model for controlling and adjusting the behavior of the model; the internal parameters of each fusion model include the learning rate, weight coefficient, bias term, activation function parameter, regularization coefficient and error tolerance of the fusion model.

[0121] S4.4: Based on the set of fusion models after weight adjustment and parameter optimization, re-evaluate the current environmental state, and build a standardized control execution instruction set containing control targets, action devices, adjustment directions and execution amplitudes.

[0122] Specifically, based on the current environmental state evaluation results in the set of fusion models after weight adjustment and parameter optimization, the related parameters of the current environmental state are re-acquired, the latest evaluation results of the current environmental state are generated, and based on the latest evaluation results of the current environmental state, the control variable attribution analysis method is used to analyze the control target, action device and adjustment direction. Based on the latest evaluation results of the current environmental state, by keeping other environmental variables unchanged, the influence of a single control variable (such as the operating parameters of a certain ventilation device) on the change of environmental state index is observed one by one, the degree of influence of the control variable on the adjustment of environmental state is evaluated, and the sensitivity and direction of action of each device on the adjustment of environmental state are judged by combining the operating capacity and parameter limit of the ventilation device, the execution amplitude of each control device is determined, the device control parameters are acquired, and then according to the device control parameters, the control target and the corresponding action device, adjustment direction and execution amplitude are organized and coded according to the preset control instruction format to generate a standardized control execution instruction containing control targets, action devices, adjustment directions and execution amplitudes. Finally, according to the matching relationship between each control target and execution device, the standardized control execution instructions are integrated to generate a set of standardized control execution instructions containing control targets, action devices, adjustment directions and execution amplitudes.

[0123] It needs to be explained that the control variable attribution analysis method is an analysis technique for identifying and quantifying the influence of each control variable on the target index. The control variable attribution analysis method determines the contribution size and direction of each control variable in achieving the control target by comparing the response effect of different control variables on the environmental state, thereby providing a scientific basis for the formulation of device control parameters.

[0124] S5: Adjusting the ventilation operation parameters of the plurality of sensing devices in the tunnel according to the standardized control execution instruction set to generate an optimized ventilation operation parameter set.

[0125] Specifically, the steps are as follows,

[0126] S5.1: Extracting and mapping the operation parameter type required to be adjusted for each ventilation device according to the target state value interval in the standardized control execution instruction set, constructing the correspondence between the device parameter and the target state interval as the device parameter target state mapping table.

[0127] Specifically, by analyzing the target state information contained in the control instruction, the operation parameter type required to be adjusted for each ventilation device is extracted, and a mapping rule is used to associate and map the extracted operation parameter type and the corresponding target state value interval, to construct the correspondence between the device parameter and the target state interval as the complete device parameter target state mapping table.

[0128] It needs to be explained that the mapping rule refers to a specific method for establishing the correspondence between the operation parameter type and the target state value interval based on the preset response relationship between the ventilation device operation parameter and the tunnel environmental state index. The mapping rule is usually determined according to the operation mechanism of the ventilation device and the environmental regulation efficiency, for example: when the target state is a temperature interval, the air speed parameter is preferentially selected for adjustment; when the target state is a gas concentration interval, the air volume parameter is preferentially matched for adjustment, thereby determining the corresponding path between the parameter type and the target state.

[0129] Among them, the preset ventilation device operation parameter is set comprehensively according to the historical operation data, the device performance specification and the tunnel environmental regulation experience, and the preset ventilation device operation parameter specifically includes the key parameter types such as the adjustable air speed, air volume and operation power of the ventilation device; The setting process of the preset ventilation device operation parameter is to collect the operation records of the ventilation device under different environmental conditions, extract the response relationship of the ventilation device to the tunnel temperature, humidity, harmful gas concentration and other indexes in the actual regulation process, and optimize the setting value of the ventilation device operation parameter through multiple rounds of simulation and on-site debugging verification.

[0130] It needs to be explained that the corresponding relationship between the device parameters and the target state interval refers to the functional response relationship between the device operating parameters (such as air volume, air speed, opening angle, operating time, etc.) and the expected environmental state target (such as temperature, carbon monoxide concentration, humidity, etc.) value interval according to the adjustment ability and environmental influence characteristics of various ventilation equipment in the tunnel ventilation system.

[0131] S5.2: According to the device parameter target state mapping table, combined with the current collected operating parameter values of each ventilation equipment, the deviation values between each parameter and the target state interval are calculated, and a device parameter adjustment reference deviation set is generated.

[0132] Specifically, the target state interval corresponding to each operating parameter type that needs to be adjusted is extracted from the device parameter target state mapping table, and the current operating parameter values of each ventilation equipment are collected. The median position is determined according to the lower limit value and the upper limit value of the target state interval. Then, the difference between the current operating parameter value and the median position is taken as the deviation, and finally all the deviation values of the operating parameters are integrated into a structured set as the device parameter adjustment reference deviation set.

[0133] It needs to be explained that the structured feature of the device parameter adjustment reference deviation set is that all the deviation values of the device operating parameters are organized according to the preset specification format, each deviation value is clearly labeled with the corresponding operating parameter type name, unit of measurement and specific deviation value, and the deviation values of different types of operating parameters are stored and managed using a unified data structure.

[0134] It needs to be explained that the lower limit value and the upper limit value of the target state interval are determined according to the safety range setting of various environmental state indicators in the tunnel environment control strategy. These settings usually come from industry specifications, safety standards, historical operating data statistics and artificial experience evaluation results.

[0135] S5.3: According to the device parameter adjustment reference deviation set, a target approaching oriented control adjustment strategy is adopted to determine the specific adjustment direction and adjustment amplitude of each parameter, and a target adjustment action sequence of the device operating parameters is generated.

[0136] Specifically, based on the positive and negative signs of the deviation values in the device parameter adjustment reference deviation set, the adjustment position of each operating parameter is determined. When the deviation value is positive, the adjustment direction is to reduce the operating parameter value, and when the deviation value is negative, the adjustment direction is to increase the operating parameter value. The specific adjustment amplitude is determined based on the absolute value of the deviation value. Then the adjustment direction and the adjustment amplitude are bound in a structured data adjustment instruction item to generate an adjustment action item containing the adjustment direction and the adjustment amplitude. Finally, all the adjustment action items of the operating parameters are arranged in order of parameter type to form the target adjustment action sequence of the device operating parameters.

[0137] It needs to be explained that the structured data adjustment instruction item refers to the standardized data structure generated for each operating parameter, and the structured data adjustment instruction item contains three fixed fields: the operating parameter type name field records the specific parameter category, the adjustment direction field explicitly labels the increase or decrease operation with a text value, and the adjustment amplitude field records the numerical change quantity with units, and the three fields form a complete parameter control instruction adjustment instruction item.

[0138] S5.4: According to the target adjustment action sequence of the device operating parameter, the adjustment instruction is issued to the corresponding ventilation equipment control interface, and the device response state is collected, and the optimized ventilation operating parameter set is generated.

[0139] Specifically, each adjustment action item in the target adjustment action sequence of the device operating parameter is analyzed item by item, the operating parameter type name, the adjustment direction and the adjustment amplitude are extracted, and the corresponding ventilation equipment control interface is matched according to the operating parameter type name, the adjustment direction and the adjustment amplitude are converted into a standard instruction format, and then the standardized adjustment instruction is issued to the physical ventilation equipment execution terminal corresponding to the matched ventilation equipment control interface through the matched ventilation equipment control interface, the response state data of the ventilation equipment after execution is collected in real time, and finally the collected response state data is integrated into a structured data set according to the parameter type, and the optimized ventilation operating parameter set is generated.

[0140] It needs to be explained that the standard instruction format includes three core fields: the operation instruction field explicitly indicates the device action type, the numerical parameter field records the signed adjustment amplitude value, and the unit identification field labels the measurement unit, and the three fields are combined into a complete instruction string through a preset delimiter.

[0141] In summary, the application realizes the dynamic optimization control of the ventilation parameter by: based on the perception device running feedback data, the feedback adjustment fusion model weight and parameter step are controlled in real time, the model weight and internal parameter are fine-tuned according to the device state feedback data, the closed-loop optimization mechanism is established, and the fault tolerance and online learning ability of the model to local anomalies are enhanced. The ventilation control system continuously outputs accurate environmental state evaluation results and standardized control instructions, and realizes the dynamic optimization control of the ventilation parameter.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalent, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for optimizing intelligent ventilation control in tunnels based on data fusion, characterized in that: include, Collect tunnel environment data from multiple sensing devices inside the tunnel, and perform noise-resistant sparse coding processing on the tunnel environment data to obtain a stable fused environmental sensing data set. The stable fusion environmental perception data set is loaded into a preset fusion model pool for fusion inference and state calculation to obtain the environmental state fusion data set; The weight allocation of each fusion model in the fusion model pool is dynamically adjusted based on the environmental state fusion data set to obtain the environmental state assessment result set. The steps are as follows: By utilizing historical tunnel operation data and manually reviewed annotation results, typical response intervals reflecting the matching relationship between environmental conditions and control commands are extracted and used as the performance evaluation benchmark for the fusion model. Based on each fusion model in the fusion model pool, the error deviation between the fusion model evaluation value and the standard state reference value is calculated, and the error deviation is mapped to the relative confidence value of each fusion model. Based on the update strategy of exponentially weighted moving average, the relative confidence value of each fusion model is dynamically adjusted, and all relative confidence values ​​are normalized to generate a dynamic set of participating weights for each fusion model under the current working condition. Based on the dynamic participation weight set of each fusion model under the current operating conditions, the environmental state estimation results of each fusion model in the fusion model pool under the current operating conditions are weighted and summarized to generate a set of environmental state assessment results under the collaboration of multiple fusion models. Based on the operational feedback data from sensing devices and combined with the environmental status assessment results, the weights and parameters of the fusion model are adjusted through real-time control feedback to generate a standardized set of control execution commands. The steps are as follows. The feedback data from the operation of sensing devices is synchronized and matched with the environmental status assessment results in time to construct a feedback dataset based on changes in the status of sensing devices and responses to environmental conditions. Based on the feedback dataset of changes in the state of sensing devices and responses to environmental conditions, the adaptability of each fusion model in the fusion model pool is determined, and the participation weight of each fusion model is dynamically adjusted according to the adaptability to obtain the set of participation weights of the fusion models that have completed dynamic weight updates. Based on the error performance of the fusion model participating in the weight set after dynamic weight updates, the internal parameters of each fusion model in the fusion model pool are fine-tuned in a targeted manner to obtain the fusion model set after weight adjustment and parameter optimization. Based on the fusion model set after weight adjustment and parameter optimization, the current environmental state is reassessed, and a standardized set of control execution instructions is constructed. The ventilation operation parameters of multiple sensing devices in the tunnel are adjusted based on a standardized set of control execution instructions, resulting in an optimized set of ventilation operation parameters.

2. The tunnel intelligent ventilation control optimization method based on data fusion as described in claim 1, characterized in that: The process involves collecting tunnel environmental data from multiple sensing devices within the tunnel, performing noise-resistant sparse coding on the tunnel environmental data, and obtaining a stable fused environmental sensing data set. The steps are as follows: Collect multidimensional raw environmental parameter data, and perform time alignment processing on the multidimensional raw environmental parameter data to generate a set of raw multidimensional tunnel environmental parameter data after time alignment; The original multidimensional tunnel environment parameter data set after time alignment is divided according to the set time window length to generate multiple environment data processing windows containing complete and continuous time series data. Based on the environmental data processing window, a time series feature vector is constructed from the original multidimensional tunnel environmental parameter data set, and a sparse coding method is used to perform sparse modeling in the preset feature base representation space to generate a multidimensional tunnel environmental parameter data set after sparse representation reconstruction. The difference between the multidimensional tunnel environment parameter data set reconstructed by sparse representation and the original multidimensional tunnel environment parameter data is calculated. Data with a difference amplitude exceeding the preset range are identified as distorted data segments and removed. Then, the sparse representation data after removing distorted points is normalized to generate a stable fused environmental perception data set.

3. The tunnel intelligent ventilation control optimization method based on data fusion as described in claim 2, characterized in that: The multidimensional raw environmental parameter data includes gas concentration parameters, aerodynamic parameters, thermal and temperature environmental parameters, traffic flow parameters, and background environmental parameters.

4. The tunnel intelligent ventilation control optimization method based on data fusion as described in claim 1, characterized in that: The steps for loading the stable fused environmental perception data set into a preset fusion model pool for fusion inference and state calculation to obtain the environmental state fused data set are as follows: Substitute the stable fusion environmental perception data set into the preset fusion model pool, perform multi-dimensional independent environmental state estimation operation, and generate an independently estimated environmental state data set; The independently estimated environmental state data sets are collected in a structured manner to construct a structured basic environmental state estimation data set; According to the preset fusion strategy, the structured basic environmental state estimation data set is multidimensionally weighted and integrated to generate the final environmental state fusion data set for feedback regulation.

5. The tunnel intelligent ventilation control optimization method based on data fusion as described in claim 4, characterized in that: The preset fusion model pool refers to a set of independent mathematical models constructed based on the environmental parameters and actual working conditions of the tunnel environment.

6. The tunnel intelligent ventilation control optimization method based on data fusion as described in claim 1, characterized in that: The steps for adjusting the ventilation operation parameters of multiple sensing devices within the tunnel based on a standardized set of control execution instructions to generate an optimized set of ventilation operation parameters are as follows: Based on the target state value range in the standardized control execution instruction set, the types of operating parameters that need to be adjusted for each ventilation device are extracted and mapped, and the correspondence between device parameters and target state ranges is constructed as a target state mapping table for device parameters. Based on the equipment parameter target state mapping table, and combined with the currently collected operating parameter values ​​of each ventilation device, the deviation values ​​between each parameter and the target state range are calculated, and a set of equipment parameter adjustment reference deviations is generated. Based on the set of reference deviations adjusted according to equipment parameters, a control adjustment strategy is adopted to determine the specific adjustment direction and adjustment range of each parameter, and a target adjustment action sequence for equipment operating parameters is generated. Based on the target adjustment sequence of equipment operating parameters, adjustment commands are issued to the corresponding ventilation equipment control interfaces, and the equipment response status is collected to generate an optimized set of ventilation operating parameters.

Images