A road sweeping optimization method, system, device and medium based on road dust dynamic prediction data

Abstract

The application relates to a road sweeping optimization method, system, device and medium based on road dust dynamic prediction data. The method comprises the following steps: constructing a sweeping effect attenuation parameter set according to road historical operation data; constructing a natural evolution benchmark model according to real-time monitoring data to generate a natural evolution parameter set; coupling the sweeping effect attenuation parameter and the natural evolution parameter to construct a state space model; and solving a future prediction time domain sweeping operation scheme based on the coupling model in combination with a road sweeping optimization target and a road sweeping constraint condition. The method can realize the transition from passive response to current pollution to active prediction of future evolution, and effectively solve the technical problems of blind selection of sweeping time and low resource utilization efficiency caused by the mutual separation of prediction and decision in the traditional method.

Description

Technical Field

[0001] This invention belongs to the field of road sweeping optimization technology, and in particular relates to a road sweeping optimization method, system, equipment and medium based on dynamic prediction data of road dust. Background Technology

[0002] With the increasing demand for refined urban air pollution control, road dust management has become an important direction in the field of environmental governance. In recent years, mobile dust monitoring networks have been widely used in buses, taxis, and other vehicles, enabling real-time collection of high-resolution road dust concentration data, providing an unprecedented sensing foundation for sanitation operations. Based on this data, some cities have begun to try to dynamically adjust cleaning operations according to real-time dust monitoring results, attempting to change the traditional extensive model of fixed-time and fixed-location cleaning and improve the utilization efficiency of cleaning resources.

[0003] In traditional technologies, road sweeping operation plans are mainly based on fixed schedules or triggered by current dust concentration thresholds. When dust levels exceed the standard on a certain road section, the dispatch center issues sweeping instructions to nearby vehicles; or, at fixed times each day, watering or vacuuming operations are carried out on all road sections at the same frequency. This model relies solely on current or historical monitoring data for decision-making, and the allocation of sweeping resources and the timing of execution lack forward-looking consideration of future dust trends.

[0004] However, this approach of responding based on the current state has a fundamental flaw: after the sweeping operation, the dust conditions on the road surface change significantly, and this change is not factored into the decision-making process. Decision-makers cannot predict how long the sweeping will take to control the concentration within the target range, nor can they determine whether sweeping at this moment is the optimal time. Due to the lack of dynamic prediction capabilities regarding the effects of sweeping intervention, existing methods struggle to achieve an optimal match between sweeping resources and dust change trends, resulting in over-sweeping in some sections and delayed response in others. Summary of the Invention

[0005] Therefore, it is necessary to provide a road cleaning optimization method that can collaboratively model the cleaning operation effect and the dynamic evolution law of dust, so as to realize the transformation from responding to the present to predicting the future.

[0006] Firstly, this application provides a road sweeping optimization method based on dynamic prediction data of road dust, including:

[0007] Based on the time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events in historical road operation data, a set of sweeping effect attenuation parameters is constructed. The sweeping effect attenuation parameter set includes the effect attenuation function for each sweeping operation type and the meteorological and traffic correction coefficient for each sweeping operation type. The sweeping operation types include watering operations, vacuum sweeping operations, and combined operations.

[0008] Based on real-time road dust concentration data, meteorological monitoring data, and traffic flow detection data, a natural evolution baseline model is constructed, and a natural evolution parameter set is generated based on the numerical parameters of the natural evolution baseline model; among them, the natural evolution baseline model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention.

[0009] Based on the set of sweeping effect decay parameters and the set of natural evolution parameters, a set of state transition equations is constructed, and a state space model is obtained based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road surface dust load state variables;

[0010] Based on the state-space model, and combining the road cleaning optimization objective and road cleaning constraints, a cleaning operation plan is generated.

[0011] Furthermore, the method also includes:

[0012] After executing the cleaning operation plan, obtain the deviation between the dust concentration data after execution and the predicted concentration trajectory of the state space model;

[0013] Based on the bias, the Bayesian sequential Monte Carlo method is used to update the key parameters in the set of cleaning effect attenuation parameters online to obtain the updated parameters; among them, the key parameters are the core physical quantities used to describe the intervention effect of the cleaning operation.

[0014] The updated parameters are substituted into the state-space model to obtain the updated state-space model, which is then used for the next round of prediction and optimization.

[0015] Furthermore, based on time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events from historical road operation data, a set of sweeping effect attenuation parameters is constructed, including:

[0016] The system retrieves time-series data on dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events for each road segment from the historical database. Based on these data, isolated operation event samples that meet preset conditions are selected.

[0017] For each isolated operation event sample, the dust concentration time series data of the preset time period before the operation is fitted to obtain the fitting result, and based on the fitting result, the theoretical concentration at each time after the operation if there is no cleaning operation is extrapolated.

[0018] Based on the actual observed concentration and the theoretical concentration, the actual cleaning effect value at each time point is calculated, and all the actual cleaning effect values ​​are arranged in chronological order to obtain the time-series effect sequence.

[0019] Based on the time-series effect sequence and corresponding sample information, different functions are used to fit the data according to the type of cleaning operation, so as to obtain the effect decay function and the baseline parameters of the effect decay function for each isolated operation event sample under the baseline meteorological and traffic conditions. The corresponding sample information includes the type of cleaning operation, operation duration, wind speed during operation and traffic flow data during operation.

[0020] By performing joint regression analysis on the baseline parameters of the effect decay function of the same type of sweeping operation, wind speed during the operation period, and traffic flow data during the operation period, a meteorological traffic correction function is generated. Based on the effect decay function, the baseline parameters of the effect decay function, the meteorological traffic correction function, and the parameters of the meteorological traffic correction function, a set of sweeping effect decay parameters is generated.

[0021] Furthermore, based on the set of cleaning effect attenuation parameters and the set of natural evolution parameters, a set of state transition equations is constructed, and based on the set of state transition equations, a state-space model is obtained, including:

[0022] Set the state vector of each road segment at discrete time, and construct the mass conservation relationship between the state variables in the state vector;

[0023] Based on the mass conservation relationship and the mass balance equation parameters in the natural evolution parameter set, a discrete equation for the natural evolution of the state without cleaning intervention is constructed; among which, the discrete equation for the natural evolution of the state includes the concentration evolution equation and the dust load evolution equation;

[0024] Obtain the effect decay function of each cleaning operation type corresponding to the current road segment from the set of cleaning effect decay parameters, and transform each effect decay function into an intervention expression for the state variable;

[0025] By combining the discrete equations of natural state evolution and the intervention expression, a set of state transition equations is obtained, and based on the set of state transition equations, a state-space model is obtained.

[0026] Furthermore, based on the state-space model, and combining the road cleaning optimization objective and road cleaning optimization constraints, a cleaning operation plan is generated, including:

[0027] At each decision moment, based on the state vector and state space model at the current moment, the concentration evolution trajectory of each road segment within the first preset future time domain length under the condition of no new sweeping intervention is predicted in a rolling manner to obtain the predicted trajectory, and the predicted trajectory is determined as the baseline prediction sequence.

[0028] A road cleaning optimization objective function is constructed, with the cleaning motion of each road segment within a second preset future time domain as the decision variable. The expression of the road cleaning optimization objective function is as follows:

[0029]

[0030] in, Optimize the objective function for road cleaning. The target value for dust concentration control. For concentration penalty weights, Dust load penalty weight, This is the cost coefficient for watering operations. This is the cost coefficient for vacuuming and sweeping operations. and It is a binary variable, and a value of 1 indicates that at time 1... Perform the corresponding type of cleaning operation. , For the action of sprinkling water, For the suction and sweeping action, C represents the dust load, and C represents the dust concentration. The first preset future time domain length, The second preset future time domain length;

[0031] To optimize the objective function of road cleaning, constraints are set for road cleaning, generating a constrained optimization problem. These constraints include state transition equation constraints, state boundary constraints, mutual exclusion constraints for actions, minimum time interval constraints, a limit on the total number of available cleaning vehicles, and a limit on the maximum duration of a single operation for a single vehicle.

[0032] Based on the baseline prediction sequence, a constrained optimization problem is solved to generate a cleaning operation plan.

[0033] Furthermore, based on the baseline prediction sequence, a constrained optimization problem is solved to generate a cleaning operation plan, including:

[0034] Based on the baseline prediction sequence, the marginal benefits of sweeping actions at each time point of each road segment are calculated by solving the adjoint equation. Based on the marginal benefits, sweeping vehicles and operation time periods are allocated to obtain the upper-level resource allocation scheme.

[0035] Based on the upper-level resource allocation scheme, the dynamic programming method is used to solve the optimal cleaning time and optimal operation type of a single road segment, and based on the optimal cleaning time and optimal operation type, the lower-level trajectory optimization scheme is generated.

[0036] Calculate the new marginal benefit, and based on the new marginal benefit, repeat the steps from generating the upper-level resource allocation scheme to generating the lower-level trajectory optimization scheme until the change in the road sweeping optimization objective function between two adjacent iterations is less than a preset threshold, then stop the iteration and obtain the optimal lower-level trajectory optimization scheme for all road segments.

[0037] By summarizing all the optimal lower-level trajectory optimization schemes, a cleaning operation plan is obtained.

[0038] Secondly, this application also provides a road sweeping optimization system based on dynamic prediction data of road dust, including:

[0039] The dataset construction module is used to construct a set of sweeping effect attenuation parameters based on time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events from historical road operation data. The sweeping effect attenuation parameter set includes the effect attenuation function for each sweeping operation type and the meteorological and traffic correction coefficient for each sweeping operation type. The sweeping operation types include watering operations, vacuum sweeping operations, and combined operations.

[0040] The parameter acquisition module is used to construct a natural evolution baseline model based on real-time road dust concentration data, meteorological monitoring data, and traffic flow detection data, and to generate a natural evolution parameter set based on the numerical parameters of the natural evolution baseline model; wherein, the natural evolution baseline model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention.

[0041] The model generation module is used to construct a set of state transition equations based on the sweeping effect decay parameter set and the natural evolution parameter set, and to obtain a state space model based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road dust load state variables;

[0042] The scheme generation module is used to generate cleaning operation schemes based on the state space model, combined with the road cleaning optimization objectives and road cleaning constraints.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement any of the road sweeping optimization methods based on dynamic prediction data of road dust as described in the embodiments of this application.

[0044] Fourthly, this application also provides a computer-readable storage medium storing at least one piece of program code, which is loaded and executed by a processor to implement a road sweeping optimization method based on dynamic prediction data of road dust as described in any of the embodiments of this application.

[0045] The aforementioned road sweeping optimization method, system, equipment, and medium based on dynamic prediction data of road dust, constructs a set of sweeping effect attenuation parameters based on historical road operation data; generates a set of natural evolution parameters based on real-time monitoring data; couples the sweeping effect attenuation parameters and natural evolution parameters to construct a state-space model; and generates a sweeping operation plan based on this coupled model. This enables a shift from passively responding to current pollution to proactively predicting future evolution, effectively solving the technical problems of traditional methods where prediction and decision-making are disconnected, leading to blind selection of sweeping timing and low resource utilization efficiency. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating a road sweeping optimization method based on dynamic prediction data of road dust in one embodiment;

[0048] Figure 2 This is a flowchart illustrating the steps of constructing a set of cleaning effect attenuation parameters based on time-series data of dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation event records from historical road operation data in one embodiment.

[0049] Figure 3 This is a schematic diagram of a road sweeping optimization system based on dynamic prediction data of road dust in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] In one embodiment, a road sweeping optimization method based on dynamic prediction data of road dust is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. Figure 1 As shown, in this embodiment, the method includes the following steps:

[0052] Step S101: Based on the time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events in the historical road operation data, construct a sweeping effect attenuation parameter set; wherein, the sweeping effect attenuation parameter set includes the effect attenuation function of each sweeping operation type and the meteorological and traffic correction coefficient of each sweeping operation type; sweeping operation types include watering operation, vacuum sweeping operation, and combined operation.

[0053] For example, time-series data of dust concentration recorded for each road segment in the past period can be extracted from a historical database. Simultaneously, time-series data of meteorological parameters and traffic flow parameters for the corresponding period can be obtained. Furthermore, cleaning operation event records, including event timestamps and operation durations for watering operations, vacuuming operations, and combined watering and vacuuming operations, can be read from the sanitation operation management system. Based on this data, a cleaning effect attenuation parameter set is constructed. This parameter set includes the effect attenuation function expressions and their correction coefficients for each cleaning operation type under different meteorological and traffic conditions. Here, the historical database refers to a database specifically storing past data, primarily used for querying and analysis; the sanitation operation management system refers to an information management platform used for planning, scheduling, monitoring, and evaluating urban environmental sanitation operations such as road cleaning and watering for dust suppression; and a road segment refers to a continuous section of a road with a clearly defined start and end point, representing a local interval divided from the entire road.

[0054] Step S102: Based on real-time road dust concentration data, meteorological monitoring data, and traffic flow detection data, a natural evolution baseline model is constructed, and a natural evolution parameter set is generated based on the numerical parameters of the natural evolution baseline model; wherein, the natural evolution baseline model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention.

[0055] For example, based on real-time collected road dust concentration data, meteorological monitoring data, and traffic flow detection data, a natural evolution baseline model is constructed by integrating mechanism modeling and data-driven methods: First, a continuous mechanism model differential equation for road dust concentration is established based on the principle of matter conservation. This differential equation includes traffic emission source terms, dust resuspension source terms, dry deposition sink terms, and diffusion dilution sink terms, and its specific form can be: Unknown parameters such as the resuspension coefficient and settling velocity in the differential equation can be determined by fitting historical data, generating initial mechanistic model parameters. The differential equation of the continuous mechanistic model is then discretized to obtain discrete state transition equations with a fixed time step. .in The discrete time step is fixed, which can be set to five minutes. All input quantities on the right-hand side of the equation are taken from the current time. The numerical value. This discrete equation is the discrete mechanism model, based on the dust concentration at the current moment. Traffic emissions Resuspension item Wind speed Based on the data, the predicted concentration value for the next time step can be calculated directly. By repeatedly applying this recursive relationship at each time step, the concentration prediction sequence for any future time can be obtained. A Long Short-Term Memory (LSTM) network is constructed as a residual correction module. The input features of this network include the lag value of the mechanistic model prediction error at the current time, wind speed, wind direction, humidity, precipitation, traffic flow, heavy vehicle ratio, hourly code, and road surface wetness index. The network output is the residual correction term for the concentration prediction at the next time step. The concentration predicted by the mechanistic model and the residual correction term of the neural network are added together to obtain the natural evolution predicted concentration. A joint training method can be used to first fix the mechanistic model parameters and train the neural network to fit the residuals. Then, the neural network is embedded into the overall model for fine-tuning. After fine-tuning, a natural evolution baseline model is obtained. The optimized mechanistic model parameters and neural network weights constitute a natural evolution parameter set. This parameter set is used to fully describe the quantitative law of dust concentration evolution with meteorological, traffic, and other factors under conditions without cleaning intervention.

[0056] in, This is a direct emission item related to traffic volume and vehicle type, and can be estimated using the following formula: , For vehicle type indexing, The number of vehicles passing through per unit time. The emission factor for a single vehicle can be calculated and determined based on localized measured data. For the length of the road segment, The mixing layer height, in meters, represents the extent of vertical atmospheric mixing and is used to convert concentration into the total amount within a unit area column. The resuspension of accumulated dust depends on the road dust load and the intensity of vehicle disturbance; road dust load It is a dynamic variable, controlled by settling input and sweeping removal, and the resuspension term can be parameterized as a product of traffic flow and road dust load: ; The resuspension coefficient can be obtained through joint inversion of dust load sampling and concentration observation; joint inversion refers to using observation data from two or more different physical fields simultaneously to obtain a unified underground geological / physical property model; road dust load A separate equation of state needs to be established and coupled with the concentration equation for solution. Coupled solution means that multiple physical quantities / equations that influence and depend on each other cannot be calculated separately. They must be set together and solved simultaneously to obtain a set of self-consistent results. This is a dry settlement sink term used to describe the process of particulate matter deposition onto the road surface; The settling velocity depends on factors such as particle size, road surface roughness, and atmospheric stability, and can be calculated using the Sehmel-Hodgson model. The Sehmel-Hodgson model is a classic parameterized model for particulate matter dry settling velocity in atmospheric science. Its core is to calculate the dry settling rate of particulate matter from the atmosphere to the surface using particle size, meteorological conditions, surface conditions, and particulate matter properties. This is a diffusion-dilution term used to describe the concentration reduction effect of horizontal transport and vertical mixing; For horizontal wind speed, The characteristic length of the road segment; The clamping speed is used to characterize the vertical mixing intensity; mechanistic modeling refers to modeling based on physical laws, formulas, and prior knowledge, which is interpretable and has strong generalization, but it is difficult to fully describe complex / unknown systems; data-driven modeling refers to modeling based on data and statistical / machine learning, which has strong fitting ability and good adaptability, but it depends on data quality, has weak interpretability, and is prone to overfitting; the fusion of mechanistic modeling and data-driven methods refers to using mechanistic models to provide structural constraints and prior knowledge, and using data-driven methods to correct errors and fit unknown parts, thus taking the best of both worlds; the principle of conservation of matter, also known as the principle of mass conservation, states that in a closed, isolated system, matter cannot be created or destroyed out of thin air, but only transforms from one form to another, while the total mass remains unchanged; Long Short-Term Memory (LSTM) networks are a type of... The unique recurrent neural network architecture aims to solve the gradient vanishing or gradient explosion problems faced by traditional recurrent neural networks when processing long sequences. By introducing gating mechanisms including forget gates, input gates, and output gates, it selectively retains, updates, or forgets information in the sequence, thereby effectively modeling long-term dependencies. Joint training refers to multiple participants training the same model together without directly exchanging raw data. Fitting refers to using a function / curve / model to closely approximate a set of existing data to describe patterns and predict the unknown. Embedding refers to treating it as a module and piecing it into the complete task process. Fine-tuning refers to further training the pre-trained model using a small amount of labeled data from a specific domain or task to better adapt the model to the specific application scenario.

[0057] Step S103: Based on the sweeping effect attenuation parameter set and the natural evolution parameter set, construct a set of state transition equations, and obtain a state space model based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road surface dust load state variables.

[0058] For example, based on the effect decay function in the cleaning effect decay parameter set and the optimized mechanism model parameters in the natural evolution parameter set, a complete set of state transition equations is constructed, and the set of equations is organized into the standard form of a state-space model, including a state transition matrix, a cleaning action influence matrix, and nonlinear coupling terms. At the same time, observation equations are established to map state variables to concentration observations from mobile monitoring and fixed stations, and finally a state-space model that can reflect the mechanism by which cleaning intervention affects state variables is generated.

[0059] Step S104: Based on the state-space model, and combining the road cleaning optimization objective and road cleaning constraints, a cleaning operation plan is generated.

[0060] For example, using a state-space model as the prediction core and minimizing dust concentration and sweeping operation costs as the road sweeping optimization objectives, under constraints on the number of sweeping vehicles, operation time intervals, and total resources, a rolling time-domain optimization method can be used to solve for the optimal timing and path of sweeping operations in the future prediction time domain. Based on this optimal timing and path, a sweeping operation plan including the operation type, operation time, and designated vehicle number for each road segment is generated. The rolling time-domain optimization method refers to a dynamic optimization method that breaks down long-cycle global optimization into a series of finite-time-domain local optimizations, solves them in a rolling fashion, executes only the current step, and provides real-time feedback correction.

[0061] In this embodiment, a set of sweeping effect attenuation parameters is constructed based on historical road operation data; a set of natural evolution parameters is constructed based on real-time monitoring data; the sweeping effect attenuation parameters and natural evolution parameters are coupled to construct a state-space model; and a sweeping operation plan is generated by rolling time-domain optimization based on this coupled model. This approach incorporates the sweeping operation effect as an endogenous variable of the prediction model into forward-looking decision-making, enabling the generated sweeping plan to predict the future differences in the effects of sweeping operations performed at different times. This achieves a shift from passively responding to current pollution to proactively predicting future evolution, effectively solving the technical problems of blindly choosing sweeping times and low resource utilization efficiency caused by the separation of prediction and decision-making in traditional methods.

[0062] In one exemplary embodiment, the method further includes:

[0063] Step S201: After executing the cleaning operation plan, obtain the deviation between the dust concentration data after execution and the predicted concentration trajectory of the state space model.

[0064] For example, after executing the cleaning operation plan, measured dust concentration data for a preset time period after execution is obtained. Simultaneously, the concentration trajectory data predicted for the corresponding road segment within the same time period using the current parameters is extracted from the state-space model. The measured concentration data and predicted concentration trajectories are aligned using the same timestamp, and the deviation value for each observation moment is calculated. This deviation is defined as the difference or relative error between the measured and predicted concentrations. These deviation values ​​are summarized to obtain a deviation sequence reflecting the difference between the model's prediction accuracy and the actual cleaning operation effect. The statistical characteristics of this deviation sequence reflect the accuracy of the current model parameters in describing the cleaning intervention effect. If the deviation is systematically large or small, it indicates that some key parameters in the set of cleaning effect attenuation parameters need adjustment. The preset time period refers to a pre-set and pre-defined time period.

[0065] Step S202: Based on the bias, the key parameters in the set of cleaning effect attenuation parameters are updated online using the Bayesian sequential Monte Carlo method to obtain the updated parameters; among them, the key parameters are the core physical quantities used to describe the intervention effect of the cleaning operation.

[0066] Among them, the Bayesian sequential Monte Carlo method, also known as particle filtering, is a sequential inference technique based on Bayesian theory and Monte Carlo simulation. Its core principle is to adjust the probability distribution of parameters according to the degree of agreement between observed data and model predictions, with the deviation being a quantitative indicator of the degree of agreement. Bayesian theory refers to updating existing judgments with new evidence. According to Bayes' theorem, the posterior distribution is proportional to the product of the likelihood function and the prior distribution. Monte Carlo simulation refers to using a large number of random samples instead of complex formula calculations to estimate results or risks. Online updates refer to dynamically modifying key data such as core configurations and model parameters directly during operation without restarting, stopping the system, or redeploying, and the changes take effect immediately. Key parameters include the attenuation time constant of water spraying operations, the ultimate efficiency of vacuum sweeping operations, and the coupling coefficient of combined operations, which are core physical quantities used to describe the intervention effect of sweeping operations.

[0067] For example, a set of weighted particles is drawn from the prior distribution, with each particle representing a candidate value for a set of key parameters. Each particle is substituted into a state-space model to obtain a predicted concentration trajectory, which is then compared with the measured concentration. The deviation value at each observation time is calculated, and an observation likelihood function is constructed based on these deviation values. Particles with smaller deviations receive higher likelihood values, indicating that the set of parameters better explains the observed sweeping effect; particles with larger deviations have lower likelihood values, indicating that the set of parameters does not match the actual situation. The weight of each particle is updated according to the likelihood value to achieve a recursive approach from the prior distribution to the posterior distribution. When particle weights degenerate, a resampling operation is performed, replicating high-weight particles and discarding low-weight particles, using the weighted expected value of the particle set as the updated parameters. The prior distribution can be pre-set based on the statistical characteristics of the fitting results of historical data. For example, it can be assumed that each key parameter follows a normal distribution, and its mean is the benchmark parameter value obtained by fitting historical samples in step S101. The variance can be set according to the fitting residual or experience, or a uniform distribution can be used as an uninformed prior when there is no prior information. Resampling refers to changing the sampling rate / resolution of a set of data. The likelihood value can be calculated based on the probability distribution of the observation error. Assuming that the observation error follows a zero-mean normal distribution, the likelihood function is expressed as the product of the probability densities of the observation error at each time step.

[0068] Step S203: Substitute the updated parameters into the state space model to obtain the updated state space model. The updated state space model is used to participate in the next round of prediction and optimization.

[0069] For example, the updated parameters are substituted into the set of cleaning effect attenuation parameters, replacing the original corresponding parameter values, to obtain the updated set of cleaning effect attenuation parameters. Furthermore, the updated set of cleaning effect attenuation parameters is combined with the set of natural evolution parameters to reconstruct the state transition equations. This reconstructs the expression for the cleaning intervention term in the updated state-space model based on the new cleaning intervention effect, generating the updated state-space model. This updated state-space model reflects the cleaning effect attenuation patterns revealed by the latest observational data, and its predictive ability is corrected and improved. The updated state-space model is used for predicting future dust concentrations and optimizing cleaning operation plans at subsequent decision-making moments.

[0070] In this embodiment, the deviation between the measured concentration after execution and the model prediction is obtained. Using the Bayesian sequential Monte Carlo method, key physical quantities in the sweeping effect attenuation parameter set are probabilistically inferred and corrected based on this deviation to obtain updated parameters. These updated parameters are then re-injected into the state-space model, resulting in an updated state-space model used for the next round of prediction and optimization. This effectively solves the technical problem of inaccurate predictions and decision-making failures caused by fixed model parameters in traditional methods, significantly improving the intelligence level and long-term operational reliability of road dust control.

[0071] In one embodiment, such as Figure 2 As shown, based on the time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events in historical road operation data, a set of sweeping effect attenuation parameters is constructed, including:

[0072] Step S301: Obtain time-series data of dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events for each road segment from the historical database, and filter out isolated operation event samples that meet preset conditions based on the time-series data of dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events.

[0073] Preset conditions refer to the premises, assumptions, or constraints that are pre-set and assumed to be true before a task, analysis, or reasoning is carried out.

[0074] For example, time-series data on dust concentration, meteorological parameters, and traffic flow parameters stored for each road segment are acquired, along with records of cleaning operation events. This data is organized by timestamp and road segment index. Based on this multi-source data, for each cleaning operation event, it is checked whether there are other cleaning operation records within at least two hours before the operation and whether there are any precipitation events within one hour after the operation, while ensuring that the dust concentration data for that period is complete and without missing data. Events that meet the above preset conditions are marked as isolated operation events. The operation time, operation type, operation duration, dust concentration time-series data for one hour before and after the operation, as well as wind speed and traffic flow data during the operation, corresponding to each isolated operation event are extracted and integrated to generate an isolated operation event sample set.

[0075] Step S302: For each isolated operation event sample, fit the time series data of dust concentration in the preset time period before the operation to obtain the fitting result, and extrapolate the theoretical concentration at each time after the operation if there is no cleaning operation.

[0076] The preset pre-task time period refers to the fixed time period before the task to be selected, which is set in advance.

[0077] For example, for each isolated operation event sample, the dust concentration time series data from 15 minutes before the start of the operation to the start time of the operation is extracted as the fitting interval. An Autoregressive Integrated Moving Average (ARIMA) model can be used to fit the concentration sequence composed of the dust concentration time series within this interval. The model order is determined by identifying the autocorrelation function and partial autocorrelation function of the sequence, and the model parameters can be solved using maximum likelihood estimation. The fitted ARIMA model is used to capture the trend and periodic characteristics of the concentration before the operation. This model is extrapolated forward to each time point after the operation to obtain the theoretical concentration value sequence without the intervention of the cleaning operation. During the extrapolation process, the model assumes that meteorological and traffic conditions maintain the same evolutionary pattern as before the operation to exclude the contribution of other factors to the concentration change. Among them, the differencing autoregressive moving average model is a classic statistical model used in time series analysis to model and predict non-stationary time series; differencing refers to performing first-order or multi-order differencing on the original series to eliminate trends or seasonality and make it stationary; autoregression indicates that the current value is linearly dependent on its lagged values ​​over several periods; moving average indicates that the current value is linearly affected by prediction errors (white noise) over several past periods; the autocorrelation function is a function used in statistics and signal processing to measure the degree of linear correlation between the values ​​of a time series or stochastic process at different time points. It describes the similarity between the signal and itself after time shift (time lag); the partial autocorrelation function is an important statistical tool in time series analysis, used to measure the degree of direct linear correlation between the time series and itself at a certain lag order after removing the influence of intermediate lags; maximum likelihood estimation refers to finding the model parameters most likely to produce the known result; forward extrapolation refers to the method of extending predictions to unknown intervals in the future / further ahead based on known data / trends.

[0078] Step S303: Based on the actual observed concentration and the theoretical concentration, calculate the actual cleaning effect value at each time point, and arrange all the actual cleaning effect values ​​in chronological order to obtain the time-series effect sequence.

[0079] For example, based on the generated theoretical concentration sequence and the post-operation concentration sequence obtained from actual observation, the actual cleaning effect value is calculated one by one at the same time points. The formula for calculating the actual cleaning effect value is: This formula reflects the proportion by which the concentration decreases due to the cleaning operation, thus eliminating the influence of the background concentration value on the effectiveness evaluation. Among other things, This represents the actual cleaning effect at time t after the operation, and is dimensionless. The actual observed concentration at time t is expressed in micrograms per cubic meter. Let be the theoretical uninterrupted concentration at time t, in micrograms per cubic meter. For example, all calculated values... Arranged chronologically, a time-series effect sequence reflecting the decay process of the effect over time is obtained. This sequence is used to show the relative decrease in dust concentration after the sweeping operation and its subsequent evolution trend.

[0080] Step S304: Based on the time-series effect sequence and corresponding sample information, different functions are used to fit the data according to the type of sweeping operation, so as to obtain the effect decay function and the baseline parameters of the effect decay function for each isolated operation event sample under the baseline meteorological and traffic conditions; wherein, the corresponding sample information includes sweeping operation type, operation duration, wind speed during operation and traffic flow data during operation.

[0081] The operation duration is used to determine the effective range of the data sequence, excluding periods when the operation itself has not yet ended; wind speed and traffic flow data during the operation do not directly participate in the curve fitting of the current sample, but are recorded as the environmental label corresponding to the sample.

[0082] For example, the time-series effect sequence and the corresponding isolated job event samples are read, along with the job type and duration. Using the time-series effect sequence as the target value, the fitting function parameters are adjusted to make the theoretical curve as close as possible to these measured points. Different fitting functions are selected based on different sweeping job types. For watering job samples, a double exponential function can be used: Nonlinear least squares fitting is performed; for suction and scanning samples, a combination function of Logistic growth and exponential decay can be used: To perform fitting; for joint operation samples, a weighted combination function of exponential decay and linear decay can be used: A fitting process is performed, and the parameters obtained from the fitting are the baseline parameters of the effect decay function for this sample under the baseline meteorological and traffic conditions. Among them, Let be the attenuation function of the watering operation effect. For amplitude coefficient, and Let these be the rise time constant and the decay time constant, respectively, and satisfy the following conditions: ; This is the start time of the task; This is the attenuation function for the suction and sweeping operation effect; To maximize the effect, For growth rate, To delay time, The decay time constant; This is the attenuation function for the combined operation effect; This is the initial effect. For exponentially decaying weights, It is the exponential decay time constant. For linear decay duration, sign This indicates the operation of taking positive values; nonlinear least squares fitting refers to approximating a set of data with a non-straight curve, minimizing the sum of squared errors of all points on the curve; Logistic growth is a typical S-shaped curve function, and its mathematical form is usually expressed as: ; This is the upper limit value. For growth rate, This is the inflection point.

[0083] Step S305: Perform joint regression analysis on the baseline parameters of the effect decay function of the same type of sweeping operation, the wind speed during the operation period, and the traffic flow data during the operation period to generate a meteorological traffic correction function. Based on the effect decay function, the baseline parameters of the effect decay function, the meteorological traffic correction function, and the parameters of the meteorological traffic correction function, generate a sweeping effect decay parameter set.

[0084] Joint regression analysis refers to a statistical analysis method that establishes regression models for multiple dependent variables simultaneously. Unlike traditional single-variable regression, it uses statistical regression to quantify the impact of wind and vehicles on cleaning into a correction formula, making the cleaning effect assessment closer to the actual road conditions.

[0085] For example, a joint regression analysis is performed on the baseline parameters of all samples under the same cleaning operation type and the corresponding wind speed and traffic flow data during the operation period to establish a correction function for the baseline parameters as a function of wind speed and traffic flow. Taking the decay time constant of watering operations as an example, the correction function can be in power law form: .in, These are parameter values ​​corrected for meteorological and traffic conditions. As the baseline parameter, For wind speed, For reference wind speed, For traffic flow, For reference traffic flow, and The wind speed correction index and traffic flow correction index are determined through regression analysis. For example, similar correction functions are established for all key baseline parameters, and the effect decay function expression, baseline parameters, correction function forms, and their parameters are integrated and stored to generate a sweeping effect decay parameter set. This parameter set is used to describe the effect decay laws of watering, vacuuming, and combined operations under different meteorological and traffic conditions. Integrated storage refers to the structured storage of the complete set of mathematical / logical rules describing how the effect changes with conditions.

[0086] In this embodiment, isolated clean operation event samples are selected from massive historical operation data; the theoretical concentration is extrapolated using the ARIMA model to calculate the time series sequence of actual cleaning effects; different function forms are used to fit different operation types to extract benchmark parameters; a meteorological and traffic correction function is established through joint regression analysis, and the effect decay function expression, benchmark parameters, correction function form and its parameters are integrated and stored to generate a cleaning effect decay parameter set. This transforms the originally empirical understanding of cleaning effects into a calculable and predictable parameterized mathematical expression, enabling subsequent optimization decisions to make forward-looking predictions based on the quantitative law of cleaning effect decay over time.

[0087] In one embodiment, a set of state transition equations is constructed based on a set of cleaning effect attenuation parameters and a set of natural evolution parameters, and a state-space model is obtained based on the set of state transition equations, including:

[0088] Step S401: Set the state vector of each road segment at discrete time and construct the mass conservation relationship between the state variables in the state vector.

[0089] For example, for each road segment to be analyzed, its state vector at a discrete time step is defined. This vector consists of two state variables: dust concentration and dust concentration. and road dust load Based on the principle of mass conservation, the intrinsic relationship between these two state variables is constructed: road dust load is resuspended into the air through vehicle disturbance, increasing dust concentration; airborne particulate matter returns to the road surface through dry settling, increasing dust load. This two-way mass exchange process constitutes the physical basis for the coupled evolution of the two variables, which can be expressed as a mass conservation relationship. This represents the mass of particulate matter per unit volume of air in that road section at time t, expressed in micrograms per cubic meter. This represents the mass of airborne particulate matter deposited per unit area of ​​the road surface at time t, expressed in grams per square meter.

[0090] Step S402: Based on the mass conservation relationship and the mass balance equation parameters in the natural evolution parameter set, construct the state natural evolution discrete equation without cleaning intervention; wherein, the state natural evolution discrete equation includes the concentration evolution equation and the dust load evolution equation.

[0091] For example, a predetermined discrete state transition equation and parameter values ​​are read from the set of natural evolution parameters and defined as the concentration evolution equation in the discrete equation of natural state evolution without cleaning intervention. Simultaneously, based on the mass conservation relationship, the dust load evolution equation in the discrete equation of natural state evolution is established: This equation reflects the effect of sedimentation flux on dust load increase and resuspension flux on dust load decrease. The mass conservation relation provides the physical basis for introducing the dust load evolution equation, allowing the two equations to be understood as a coupled system. The length of the road segment; The discrete time step; Indicates the next moment Dust load.

[0092] Step S403: Obtain the effect decay function of each cleaning operation type corresponding to the current road segment from the cleaning effect decay parameter set, and transform each effect decay function into an intervention expression for the state variable.

[0093] For example, the attenuation functions for each cleaning operation type corresponding to the current road segment are obtained from the set of cleaning effect attenuation parameters, including the attenuation function for water spraying, vacuum sweeping, and combined operation. These attenuation functions are then transformed into intervention expressions for state variables: for water spraying, its effect is to suppress the resuspension process, i.e., by modifying the resuspension coefficient, so the intervention expression can be a time-varying reduction factor for the resuspension coefficient; for vacuum sweeping, its effect is to directly remove the road dust load and introduce a short-term disturbance effect, so the intervention expression can include a deduction term for the dust load and a short-term enhancement term for the resuspension coefficient; for combined operations, the two intervention effects can be combined through a coupling coefficient.

[0094] Step S404: Combine the discrete equations of natural state evolution and the intervention expression to obtain a set of state transition equations, and based on the set of state transition equations, obtain the state space model.

[0095] For example, the discrete equations of natural state evolution and the cleanup intervention expression are combined to form a complete set of state transition equations. This set of equations uses the state vector... With cleaning as the core, the cleaning action As a control input, it can be organized into the standard form of a state-space model: .in, This is the natural evolution state transition matrix. The linear influence matrix of the cleaning action. This represents the nonlinear coupling term of the sweep intervention. For example, the observation equation is established: ,in, For concentration observation vectors collected by mobile monitoring and fixed stations, For the observation matrix, For observing noise; fixed stations refer to the deployment of monitoring equipment at fixed locations to collect data continuously over a long period of time. The data is stable and can be compared over a long period of time, but the coverage is limited; mobile monitoring refers to the use of mobile devices such as vehicles and drones to monitor different locations. The coverage is wide and flexible, but the data is discontinuous and difficult to compare over a long period of time.

[0096] In this embodiment, state variables are defined and a mass conservation relationship is established; based on the mass conservation relationship and the natural evolution parameter set, an intervention-free discrete evolution equation is constructed; an effect function is extracted from the sweeping effect decay parameter set and transformed into an intervention expression; the two are combined to form a complete set of state transition equations and a state-space model is derived. This enables the mathematical formalization of the core dynamic relationships in the "prediction-decision-execution-feedback" closed loop, allowing the model to proactively predict the concentration response under different sweeping strategies.

[0097] In one embodiment, based on a state-space model, and combining the road cleaning optimization objective and road cleaning optimization constraints, a cleaning operation plan is generated, including:

[0098] Step S501: At each decision moment, based on the state vector and state space model at the current moment, the concentration evolution trajectory of each road segment within the first preset future time domain length under the condition of no new sweeping intervention is rolled for prediction, and the predicted trajectory is determined as the baseline prediction sequence.

[0099] The first preset future time domain length is a pre-set time length, which corresponds to the time window that needs to be predicted forward; rolling prediction means using the latest data to continuously slide the window forward for prediction, rather than predicting to the end all at once.

[0100] For example, obtain the state vector of each road segment at the current moment, including dust concentration. and road dust load Simultaneously obtain the first preset future time domain length. The weather forecast data and traffic flow prediction data are used as exogenous inputs to the state-space model. The current state vector is used as the initial condition, and the exogenous input sequence is substituted into the state transition equations of the state-space model time-by-time in chronological order. This assumes future... Under the condition that no new cleaning operations are carried out within a certain time step, the state prediction values ​​at each time point are calculated recursively to obtain the concentration prediction sequence for future time points. The calculated concentration prediction sequence is associated and stored with the corresponding time label and road segment number to generate a baseline prediction sequence dataset including the prediction time and the predicted concentration value. This baseline prediction sequence is used to reflect how the dust concentration will change over time under the current natural evolution law if no new cleaning intervention is taken. Here, recursive calculation means using the previously known results to deduce the subsequent results step by step, rather than directly calculating the answer in one step; associated storage means binding the three types of information, namely the concentration prediction value (numerical sequence), time label (time dimension), and road segment number (spatial dimension), so that each concentration prediction value can be accurately mapped to a specific time and road segment.

[0101] Step S502: Construct a road cleaning optimization objective function with the cleaning motion of each road segment within a second preset future time domain length as the decision variable. The expression of the road cleaning optimization objective function is as follows:

[0102]

[0103] in, Optimize the objective function for road cleaning. The target value for dust concentration control. For concentration penalty weights, Dust load penalty weight, This is the cost coefficient for watering operations. This is the cost coefficient for vacuuming and sweeping operations. and It is a binary variable, and a value of 1 indicates that at time 1... Perform the corresponding type of cleaning operation. , For the action of sprinkling water, For the suction and sweeping action, C represents the dust load, and C represents the dust concentration. The first preset future time domain length, The second preset future time domain length.

[0104] For example, a second preset future time domain length is constructed. The objective function for road cleaning optimization is defined as follows, with the cleaning actions of each road section as the decision variables. This objective function consists of a weighted sum of three parts, and its mathematical expression is as follows: .in, The objective function value for road cleaning is optimized and is dimensionless. The preset dust concentration control target value is expressed in micrograms per cubic meter. The concentration penalty weight is used to adjust the relative importance of concentration exceeding the limit in the objective function; This is the dust load penalty weight, used to adjust the relative importance of road dust load accumulation in the objective function; and These are the cost coefficients for water spraying operations and vacuum sweeping operations, respectively. and For a binary decision variable, a value of 1 indicates that at time 1... Perform the corresponding type of cleaning job; a value of 0 indicates that the job is not performed. and They are time points The predicted values ​​of dust concentration and dust load; The first preset future time domain length is the number of steps for prediction and evaluation. The second preset future time domain length, i.e., the number of steps in decision optimization, usually satisfies .

[0105] Step S503: Set road cleaning optimization constraints for the road cleaning objective function to generate a constrained optimization problem; wherein, the road cleaning optimization constraints include state transition equation constraints, state boundary constraints, action mutual exclusion constraints, minimum time interval constraints, a limit on the total number of available cleaning vehicles, and a limit on the maximum duration of a single operation by a single vehicle.

[0106] Among these constraints, the state transition equation requires that the predicted concentration and dust load values ​​at all times strictly satisfy the established state space model; the state boundary constraints require that the concentration of each road segment must not exceed the preset safety limit at any time, and the dust load must not exceed the dust accumulation limit, to ensure environmental safety; and the mutual exclusion constraint stipulates that only one cleaning operation can be performed on the same road segment at the same time. The minimum time interval constraint requires that there must be a minimum interval between two adjacent operations of the same type. A specific time step is used to avoid over-cleaning; This refers to the minimum interval set; the total number of available sweeping vehicles is limited, stipulating that the number of tasks to be performed simultaneously shall not exceed the actual number of vehicles owned by the fleet; the maximum duration of a single operation per vehicle is limited, stipulating that the continuous operation time of each vehicle shall not exceed the set maximum duration, in order to ensure the quality of operation and equipment maintenance.

[0107] For example, by setting a series of physical realizability and resource-limited constraints on the objective function of road cleaning, the optimization problem is transformed into a constrained mathematical programming problem. These constraints collectively define the feasible region of the optimization problem, ensuring that the generated cleaning plan is executable in practice.

[0108] Step S504: Based on the baseline prediction sequence, solve the constrained optimization problem to generate a cleaning operation plan.

[0109] For example, based on the generated baseline prediction sequence, a constrained optimization problem is solved under set constraints. Since this problem involves large-scale mixed integer decision variables, a decomposition and coordination strategy can be used to solve it, generating a cleaning operation plan that includes the operation type, operation time, and designated vehicle number for each road segment. The decomposition and coordination strategy is the core methodology for handling complex / large-scale problems: first, the whole is broken down into independently solvable subproblems (decomposition), and then the solutions to the subproblems converge to the global optimum through iterative interaction (coordination).

[0110] In this embodiment, the concentration trajectory under future no-intervention conditions is predicted rollingly based on the current state and the coupled model; a multi-objective optimization function that balances dust control effectiveness and cleaning operation costs is constructed; various constraints that conform to actual operating conditions are set; and the complex optimization problem is solved through a decomposition and coordination algorithm to generate an executable optimal cleaning plan. This elevates cleaning decisions from traditional experience-based judgments or simple threshold triggers to forward-looking optimization based on dynamic prediction, enabling cleaning resources to be deployed to the most needed sections at the most needed times, maximizing cost savings while meeting environmental quality goals.

[0111] In one embodiment, based on a baseline prediction sequence, a constrained optimization problem is solved to generate a cleaning operation plan, including:

[0112] Step S601: Based on the baseline prediction sequence, the marginal benefits of sweeping actions at each time point of each road segment are calculated by solving the adjoint equation. Based on the marginal benefits, sweeping vehicles and operation time periods are allocated to obtain the upper-level resource allocation scheme.

[0113] Among them, the adjoint equation is the dual equation of the original equation. It is a core tool in optimization control theory for efficiently calculating the gradient of the objective function with respect to the control variables. Its basic principle is to introduce the adjoint state variable and transform the original optimization problem into a set of differential equations that are integrated in reverse along time. Thus, the sensitivity information of all decision variables to the objective function can be obtained in one solution.

[0114] For example, a baseline prediction sequence is used as the nominal trajectory and input into the adjoint equation solving process to calculate the marginal benefit of sweeping actions for each road segment at each time step. First, the state-space model is linearized near the nominal trajectory provided by the baseline prediction sequence, resulting in a time-varying linearized system matrix. This matrix reflects the impact of small changes in state variables on future evolution. Starting from the final time step in the prediction time domain, the adjoint equation is solved in reverse, continuously accumulating the partial derivatives of the objective function with respect to the state, ultimately propagating to obtain the marginal benefit value of each road segment at each future time step. This value quantifies the extent to which performing a cleaning action at that moment reduces the objective function value. All candidate cleaning tasks are ranked from highest to lowest marginal benefit, and vehicles and work periods are allocated sequentially based on the limited total number of available cleaning vehicles, prioritizing the road segments with the highest marginal benefit. This generates a higher-level resource allocation scheme including vehicle number, service segment, and expected arrival time window. Linearization refers to performing a first-order Taylor expansion of the nonlinear state-space model at each time point of the nominal trajectory, ignoring higher-order infinitesimal terms. This approximates the state-space model containing cleaning intervention terms as a linear time-varying system with reference to the nominal trajectory. Specifically, the Jacobian matrix of the state transition function with respect to the state variables (i.e., the partial derivative matrix of the influence of small changes in the state variables at each time point on the state at the next time point) and the partial derivative matrix of the state transition function with respect to the control variables are calculated. These partial derivative matrices constitute the time-varying linearized system matrix.

[0115] Step S602: Based on the upper-level resource allocation scheme, the dynamic programming method is used to solve for the optimal cleaning time and optimal operation type of a single road segment, and based on the optimal cleaning time and optimal operation type, a lower-level trajectory optimization scheme is generated.

[0116] Dynamic programming is a recursive optimization method based on the Bellman optimality principle. It is applicable to multi-stage decision problems. Its core idea is to decompose the original problem into a series of interrelated single-stage subproblems, calculate the optimal cost function of all possible states in each stage by recursively calculating in reverse, and then backtrack forward to obtain the optimal decision sequence. Backtracking refers to the problem-solving approach of turning back and trying a different path when a path is blocked.

[0117] For example, for a given road segment, the second preset future time domain length is... Each discrete time step within the time frame is defined as a decision stage, and the state vector of each stage is determined by the dust concentration at the current moment. and road dust load The decision variables are binary variables. and , representing whether to perform watering or sweeping operations, respectively, and are constrained by the upper-level resource allocation scheme, meaning that the decision variable is only allowed to take the value 1 during time periods when resources are allocated. Stage transitions are strictly described by a state-space model; given the current state and the decision, the state at the next moment is uniquely determined by the state transition equations. Stage costs are composed of corresponding terms in the road sweeping optimization objective function. Starting from the last stage, the optimal cost function for each possible state is calculated in reverse recursion, i.e., the minimum cumulative cost from that state to the end point, and the optimal decision that minimizes the cost is recorded. Since the state space is continuous, a discretization method can be used to divide the range of concentration and dust load values ​​into a finite number of grid points, and an approximate optimal cost function is calculated at each grid point. After the reverse recursion is completed, starting from the actual initial state at the current moment, each stage is simulated in forward motion. Based on the optimal decision stored in reverse calculation, the sweeping action to be performed in the current stage is selected, gradually advancing until the entire decision time domain is covered, thus obtaining the optimal sweeping time sequence and corresponding operation type for the road segment at future moments. The results obtained from solving each independent road segment are summarized to obtain the lower-level trajectory optimization scheme. This scheme details the cleaning action instructions for each allocated resource road segment at every moment, including the job type, job start time, job duration, and required vehicle number. The discretization method refers to mapping large / scattered numerical values ​​into a compact, continuous sequence of small integers for easier calculation and storage.

[0118] Step S603: Calculate the new marginal benefit, and based on the new marginal benefit, repeat the steps from generating the upper-level resource allocation scheme to generating the lower-level trajectory optimization scheme until the change in the road sweeping optimization objective function between two adjacent iterations is less than a preset threshold, then stop the iteration and obtain the optimal lower-level trajectory optimization scheme for all road segments.

[0119] For example, the contribution value of each road segment to the objective function after the actual implementation of the lower-level trajectory optimization scheme is extracted, that is, for each road segment... The optimal decision sequence is substituted into the part corresponding to that road segment in the road sweeping optimization objective function to calculate the contribution value of the objective function for that road segment. The actual benefit value is The overall objective function value is obtained by summing the actual benefit values ​​of all road segments. To further optimize resource allocation, marginal benefits need to be recalculated: based on the future state evolution trajectory of each road segment determined by the current lower-level scheme (obtained through simulation using a state-space model), the adjoint equation is solved again to obtain the new marginal benefits of each road segment at each time step. This new marginal benefit represents the potential impact on the objective function if an additional cleaning operation is added or removed based on the current locally optimal decision. The objective function values ​​of this iteration are then compared. and the previous iteration value If the absolute value of the change is less than the preset convergence threshold, the iteration stops; otherwise, the new marginal benefit is returned to step S601, and the upper-level resource allocation and lower-level optimization are redone until convergence. Finally, the lower-level trajectory optimization schemes of all road segments obtained in the last iteration are output as the optimal lower-level trajectory optimization schemes. The optimization scheme for each road segment includes detailed instructions for each cleaning operation planned to be executed within the future decision time domain, including the operation type, operation start time, operation duration, and vehicle number executing the task. The preset convergence threshold is a pre-defined value. When the change in the objective function value between two adjacent iterations is less than this preset convergence threshold, it indicates that resource allocation and trajectory optimization have reached an approximately optimal balance. The optimal decision sequence refers to the time sequence of cleaning action selections that minimize the cumulative cost function at each discrete time step within the second preset future time domain length for the road segment.

[0120] Step S604: Summarize all the optimal lower-level trajectory optimization schemes to obtain the cleaning operation scheme.

[0121] For example, the detailed instructions for all road segments are reorganized according to time sequence and vehicle number to generate a complete cleaning operation plan, which includes executable information such as the departure time of each cleaning vehicle, the sequence of road segments to be worked on, the expected arrival and departure times of each road segment, the travel routes between road segments, and the return time.

[0122] In this embodiment, marginal benefits are calculated and resources are initially allocated using the adjoint equation to form an upper-level scheme; dynamic programming is used to independently optimize the specific operation sequence for each road segment to form a lower-level scheme; the actual benefits of the lower-level scheme are fed back to the upper-level scheme to recalculate marginal benefits, and the global optimum is approximated through iteration; finally, an executable sweeping operation scheme is obtained. This approach effectively solves the computational complexity of large-scale mixed integer optimization problems, ensuring solution accuracy while meeting the timeliness requirements of real-time rolling optimization, enabling sweeping resources to achieve the best dust control effect at the lowest cost.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides a road sweeping optimization system based on dynamic prediction data of road dust for implementing the road sweeping optimization method based on dynamic prediction data of road dust mentioned above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of a road sweeping optimization system based on dynamic prediction data of road dust provided below can be found in the limitations of the road sweeping optimization method based on dynamic prediction data of road dust described above, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 3 As shown, a road sweeping optimization system 300 based on dynamic prediction data of road dust is provided, including:

[0126] The dataset construction module 301 is used to construct a set of sweeping effect attenuation parameters based on the time series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation event records in the historical road operation data. The sweeping effect attenuation parameter set includes the effect attenuation function of each sweeping operation type and the meteorological and traffic correction coefficient of each sweeping operation type. The sweeping operation types include watering operation, vacuum sweeping operation, and combined operation.

[0127] The parameter acquisition module 302 is used to construct a natural evolution benchmark model based on real-time road dust concentration data, meteorological monitoring data and traffic flow detection data, and generate a natural evolution parameter set based on the numerical parameters of the natural evolution benchmark model; wherein, the natural evolution benchmark model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention.

[0128] The model generation module 303 is used to construct a set of state transition equations based on the sweeping effect decay parameter set and the natural evolution parameter set, and to obtain a state space model based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road dust load state variables;

[0129] The scheme generation module 304 is used to generate a cleaning operation scheme based on the state space model, combined with the road cleaning optimization objective and the road cleaning constraints.

[0130] In one exemplary embodiment, the system further includes:

[0131] The data calculation module is used to obtain the deviation between the dust concentration data after the sweeping operation plan is executed and the predicted concentration trajectory of the state space model.

[0132] The parameter update module is used to update the key parameters in the cleaning effect attenuation parameter set online based on the deviation using the Bayesian sequential Monte Carlo method, so as to obtain the updated parameters; among them, the key parameters are the core physical quantities used to describe the intervention effect of the cleaning operation;

[0133] The model update module is used to substitute the updated parameters into the state space model to obtain the updated state space model, which is then used to participate in the next round of prediction and optimization.

[0134] In one embodiment, the dataset construction module 301 is further configured to:

[0135] The system retrieves time-series data on dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events for each road segment from the historical database. Based on these data, isolated operation event samples that meet preset conditions are selected.

[0136] For each isolated operation event sample, the dust concentration time series data of the preset time period before the operation is fitted to obtain the fitting result, and based on the fitting result, the theoretical concentration at each time after the operation if there is no cleaning operation is extrapolated.

[0137] Based on the actual observed concentration and the theoretical concentration, the actual cleaning effect value at each time point is calculated, and all the actual cleaning effect values ​​are arranged in chronological order to obtain the time-series effect sequence.

[0138] Based on the time-series effect sequence and corresponding sample information, different functions are used to fit the data according to the type of cleaning operation, so as to obtain the effect decay function and the baseline parameters of the effect decay function for each isolated operation event sample under the baseline meteorological and traffic conditions. The corresponding sample information includes the type of cleaning operation, operation duration, wind speed during operation and traffic flow data during operation.

[0139] By performing joint regression analysis on the baseline parameters of the effect decay function of the same type of sweeping operation, wind speed during the operation period, and traffic flow data during the operation period, a meteorological traffic correction function is generated. Based on the effect decay function, the baseline parameters of the effect decay function, the meteorological traffic correction function, and the parameters of the meteorological traffic correction function, a set of sweeping effect decay parameters is generated.

[0140] In one embodiment, the model generation module 303 is further configured to:

[0141] Set the state vector of each road segment at discrete time, and construct the mass conservation relationship between the state variables in the state vector;

[0142] Based on the mass conservation relationship and the mass balance equation parameters in the natural evolution parameter set, a discrete equation for the natural evolution of the state without cleaning intervention is constructed; among which, the discrete equation for the natural evolution of the state includes the concentration evolution equation and the dust load evolution equation;

[0143] Obtain the effect decay function of each cleaning operation type corresponding to the current road segment from the set of cleaning effect decay parameters, and transform each effect decay function into an intervention expression for the state variable;

[0144] By combining the discrete equations of natural state evolution and the intervention expression, a set of state transition equations is obtained, and based on the set of state transition equations, a state-space model is obtained.

[0145] In one embodiment, the scheme generation module 304 is further configured to:

[0146] At each decision moment, based on the state vector and state space model at the current moment, the concentration evolution trajectory of each road segment within the first preset future time domain length under the condition of no new sweeping intervention is predicted in a rolling manner to obtain the predicted trajectory, and the predicted trajectory is determined as the baseline prediction sequence.

[0147] A road cleaning optimization objective function is constructed, with the cleaning motion of each road segment within a second preset future time domain as the decision variable. The expression of the road cleaning optimization objective function is as follows:

[0148]

[0149] in, Optimize the objective function for road cleaning. The target value for dust concentration control. For concentration penalty weights, Dust load penalty weight, This is the cost coefficient for watering operations. This is the cost coefficient for vacuuming and sweeping operations. and It is a binary variable, and a value of 1 indicates that at time 1... Perform the corresponding type of cleaning operation. , For the action of sprinkling water, For the suction and sweeping action, C represents the dust load, and C represents the dust concentration. The first preset future time domain length, The second preset future time domain length;

[0150] To optimize the objective function of road cleaning, constraints are set for road cleaning, generating a constrained optimization problem. These constraints include state transition equation constraints, state boundary constraints, mutual exclusion constraints for actions, minimum time interval constraints, a limit on the total number of available cleaning vehicles, and a limit on the maximum duration of a single operation for a single vehicle.

[0151] Based on the baseline prediction sequence, a constrained optimization problem is solved to generate a cleaning operation plan.

[0152] In one embodiment, the scheme generation module 304 is further configured to:

[0153] Based on the baseline prediction sequence, the marginal benefits of sweeping actions at each time point of each road segment are calculated by solving the adjoint equation. Based on the marginal benefits, sweeping vehicles and operation time periods are allocated to obtain the upper-level resource allocation scheme.

[0154] Based on the upper-level resource allocation scheme, the dynamic programming method is used to solve the optimal cleaning time and optimal operation type of a single road segment, and based on the optimal cleaning time and optimal operation type, the lower-level trajectory optimization scheme is generated.

[0155] Calculate the new marginal benefit, and based on the new marginal benefit, repeat the steps from generating the upper-level resource allocation scheme to generating the lower-level trajectory optimization scheme until the change in the road sweeping optimization objective function between two adjacent iterations is less than a preset threshold, then stop the iteration and obtain the optimal lower-level trajectory optimization scheme for all road segments.

[0156] By summarizing all the optimal lower-level trajectory optimization schemes, a cleaning operation plan is obtained.

[0157] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the road sweeping optimization method based on dynamic prediction data of road dust as described above.

[0158] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0159] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0160] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A road sweeping optimization method based on dynamic prediction data of road dust, characterized in that, The method includes: Based on the time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events in historical road operation data, a sweeping effect attenuation parameter set is constructed. This set includes an effect attenuation function for each sweeping operation type and a meteorological and traffic correction coefficient for each sweeping operation type. The sweeping operation types include watering operations, vacuum sweeping operations, and combined operations. Based on real-time road dust concentration data, meteorological monitoring data, and traffic flow detection data, a natural evolution baseline model is constructed, and a natural evolution parameter set is generated based on the numerical parameters of the natural evolution baseline model; wherein, the natural evolution baseline model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention. Based on the set of sweeping effect attenuation parameters and the set of natural evolution parameters, a set of state transition equations is constructed, and a state space model is obtained based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road surface dust load state variables; Based on the state space model, and combining the road cleaning optimization objective and road cleaning constraints, a cleaning operation plan is generated.

2. The method according to claim 1, characterized in that, The method further includes: After executing the cleaning operation plan, the deviation between the dust concentration data after execution and the predicted concentration trajectory of the state space model is obtained; Based on the aforementioned deviation, the key parameters in the set of cleaning effect attenuation parameters are updated online using the Bayesian sequential Monte Carlo method to obtain the updated parameters; wherein, the key parameters are core physical quantities used to describe the intervention effect of the cleaning operation. The updated parameters are substituted into the state space model to obtain the updated state space model, which is used to participate in the next round of prediction and optimization.

3. The method according to claim 1, characterized in that, The method involves constructing a set of cleaning effect attenuation parameters based on time-series data of dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events from historical road operation data. This set includes: The system retrieves time-series data on dust concentration, meteorological parameters, traffic flow parameters, and cleaning operation events for each road segment from the historical database. Based on these data, isolated operation event samples that meet preset conditions are selected. For each isolated operation event sample, the dust concentration time series data of the preset pre-operation time period is fitted to obtain the fitting result, and based on the fitting result, the theoretical concentration at each time after the operation if there is no cleaning operation is extrapolated. Based on the actual observed concentration and the theoretical concentration, the actual cleaning effect value at each time point is calculated, and all the actual cleaning effect values ​​are arranged in chronological order to obtain a time-series effect sequence. Based on the time-series effect sequence and corresponding sample information, different functions are used to fit the cleaning operation type to obtain the effect decay function and effect decay function baseline parameters for each isolated operation event sample under the baseline meteorological and traffic conditions; wherein, the corresponding sample information includes cleaning operation type, operation duration, wind speed during operation, and traffic flow data during operation. By performing joint regression analysis on the baseline parameters of the effect decay function of the same type of sweeping operation, the wind speed during the operation, and the traffic flow data during the operation, a meteorological traffic correction function is generated. Based on the effect decay function, the baseline parameters of the effect decay function, the meteorological traffic correction function, and the parameters of the meteorological traffic correction function, a sweeping effect decay parameter set is generated.

4. The method according to claim 1, characterized in that, Based on the set of cleaning effect attenuation parameters and the set of natural evolution parameters, a set of state transition equations is constructed, and based on the set of state transition equations, a state-space model is obtained, including: Set the state vector of each road segment at discrete time, and construct the mass conservation relationship between the state variables in the state vector; Based on the mass conservation relationship and the mass balance equation parameters in the set of natural evolution parameters, a state natural evolution discrete equation is constructed without cleaning intervention; wherein, the state natural evolution discrete equation includes a concentration evolution equation and a dust load evolution equation; Obtain the effect decay function of each cleaning operation type corresponding to the current road segment from the set of cleaning effect decay parameters, and transform each effect decay function into an intervention expression for the state variable; The state natural evolution discrete equations and the intervention expression are combined to obtain the state transition equation set, and the state space model is obtained based on the state transition equation set.

5. The method according to claim 1, characterized in that, The step of generating a cleaning operation plan based on the state space model, combined with the road cleaning optimization objective and road cleaning optimization constraints, includes: At each decision moment, based on the state vector at the current moment and the state space model, the concentration evolution trajectory of each road segment within the first preset future time domain length under the condition of no new sweeping intervention is rolled and predicted to obtain the predicted trajectory, and the predicted trajectory is determined as the baseline prediction sequence. A road cleaning optimization objective function is constructed, with the cleaning actions of each road segment within a second preset future time domain length as the decision variable. The expression of the road cleaning optimization objective function is as follows: in, Optimize the objective function for road cleaning. The target value for dust concentration control. For concentration penalty weights, Dust load penalty weight, This is the cost coefficient for watering operations. This is the cost coefficient for vacuuming and sweeping operations. and It is a binary variable, and a value of 1 indicates that at time 1... Perform the corresponding type of cleaning operation. , For the action of sprinkling water, For the suction and sweeping action, C represents the dust load, and C represents the dust concentration. The first preset future time domain length, The second preset future time domain length; Set road cleaning optimization constraints for the objective function of road cleaning optimization to generate a constrained optimization problem; wherein, the road cleaning optimization constraints include state transition equation constraints, state boundary constraints, action mutual exclusion constraints, minimum time interval constraints, a limit on the total number of available cleaning vehicles, and a limit on the maximum duration of a single operation by a single vehicle; Based on the baseline prediction sequence, the constrained optimization problem is solved to generate the cleaning operation plan.

6. The method according to claim 5, characterized in that, The step of solving the constrained optimization problem based on the baseline prediction sequence to generate the cleaning operation plan includes: Based on the baseline prediction sequence, the marginal benefits of the sweeping action at each time point of each road segment are calculated by solving the adjoint equation. Based on the marginal benefits, sweeping vehicles and operation time periods are allocated to obtain the upper-level resource allocation scheme. Based on the upper-level resource allocation scheme, a dynamic programming method is used to solve for the optimal cleaning time and optimal operation type of a single road segment, and based on the optimal cleaning time and the optimal operation type, a lower-level trajectory optimization scheme is generated. Calculate the new marginal benefit, and based on the new marginal benefit, repeat the steps of generating the upper-level resource allocation scheme to generating the lower-level trajectory optimization scheme until the change in the road sweeping optimization objective function between two adjacent iterations is less than a preset threshold, then stop the iteration and obtain the optimal lower-level trajectory optimization scheme for all road segments. By summarizing all the optimal lower-level trajectory optimization schemes, the cleaning operation scheme is obtained.

7. A road sweeping optimization system based on dynamic prediction data of road dust, characterized in that, The system includes: The dataset construction module is used to construct a set of sweeping effect attenuation parameters based on time-series data of dust concentration, meteorological parameters, traffic flow parameters, and sweeping operation events in historical road operation data. The set of sweeping effect attenuation parameters includes an effect attenuation function for each sweeping operation type and a meteorological and traffic correction coefficient for each sweeping operation type. The sweeping operation types include watering operations, vacuum sweeping operations, and combined operations. The parameter acquisition module is used to construct a natural evolution benchmark model based on real-time road dust concentration data, meteorological monitoring data, and traffic flow detection data, and to generate a natural evolution parameter set based on the numerical parameters of the natural evolution benchmark model; wherein, the natural evolution benchmark model is used to describe the evolution law of dust concentration over time under conditions without cleaning intervention. The model generation module is used to construct a set of state transition equations based on the set of sweeping effect attenuation parameters and the set of natural evolution parameters, and to obtain a state space model based on the set of state transition equations; wherein, the set of state transition equations includes dust concentration state variables and road surface dust load state variables; The scheme generation module is used to generate a cleaning operation scheme based on the state space model, combined with the road cleaning optimization objective and the road cleaning constraints.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images