A sanitation operation scheduling method based on an internet of things

Abstract

The application relates to the field of sanitation operation scheduling, in particular to a sanitation operation scheduling method based on the Internet of Things, which comprises the following steps: determining an iterative stop condition of a load CEEMDAN model based on time local weighted variance of load data, so as to generate vehicle load characteristics; generating vehicle oil consumption characteristics through an oil consumption CEEMDAN model based on statistical values of load data residuals and oil consumption data residuals, so as to calculate oil consumption channel residual signals; generating comprehensive vehicle carrying characteristics through a feature extraction model based on a multilayer perception machine, a gating mechanism and a cross-modal attention mechanism; and generating corrected vehicle garbage load through a long short-term memory network and a fully connected layer architecture. The application realizes efficient noise reduction and accurate decomposition of non-stationary time series data of oil consumption and load, realizes accurate processing of Internet of Things data of garbage transfer vehicles, improves the operation efficiency of the garbage transfer vehicles and reduces the operation cost.

Description

Technical Field

[0001] This invention relates to the field of sanitation operation scheduling, and in particular to a sanitation operation scheduling method based on the Internet of Things. Background Technology

[0002] With the acceleration of urbanization, the amount of urban domestic waste generated continues to rise. As a core link in sanitation operations, waste transportation directly affects the efficiency and operating costs of urban environmental governance. As key operational equipment, the rational scheduling of waste transfer vehicles is not only related to the timeliness of waste collection and the prevention of waste accumulation and pollution, but also closely related to operating costs such as fuel consumption, equipment wear and tear, and manpower allocation. Therefore, achieving intelligent and precise scheduling of sanitation vehicles has become a core requirement in the field of urban sanitation management.

[0003] Traditional sanitation operation scheduling relies heavily on manual experience to determine fixed routes, lacking dynamic awareness of real-time vehicle capacity. This makes it impossible to adjust tasks based on the actual capacity of garbage trucks, leading to uneven capacity scheduling, idle capacity, resource waste, and reduced overall operational efficiency. Furthermore, the bumps and impacts of garbage trucks during operation cause momentary fluctuations in load sensors, making it easy to misjudge based on a single load data point. Fuel consumption data, less affected by mechanical impacts, better reflects the true load condition.

[0004] Therefore, it is currently necessary to use IoT-based deep learning algorithms to deeply mine and accurately analyze data such as fuel consumption and load during the operation process. This is to avoid relying solely on statistical data from load sensors to optimize the scheduling scheme, which has significant errors, is difficult to adapt to the dynamic needs of sanitation operation scheduling, and cannot achieve the optimal allocation of sanitation operation resources.

[0005] Furthermore, as IoT technology is gradually applied to the field of sanitation operation management, IoT devices such as fuel sensors and load sensors deployed on garbage trucks can collect real-time data on vehicle fuel consumption, load data, operation trajectory, and operating status, and transmit this information to the sanitation operation dispatch platform, providing data support for the optimization of dispatch plans.

[0006] At the data processing level of IoT devices, the fuel consumption and load data of garbage trucks are both non-stationary time-series data. They are affected by multiple factors such as road conditions, garbage type, start-stop frequency, and weather conditions, and contain a large amount of noise and fluctuation components. In existing technologies, some solutions use Empirical Mode Decomposition (EMD) models to process non-stationary data. However, traditional EMD suffers from mode aliasing problems, and it does not take into account the correlation characteristics of fuel consumption and load data in its targeted processing, resulting in insufficient data decomposition accuracy and the inability to generate reliable feature data.

[0007] Therefore, how to accurately process and integrate multi-dimensional IoT data from garbage transfer vehicles to achieve precise vehicle scheduling with correct load adjustments, thereby improving operational efficiency and optimizing costs, is a technical problem that needs to be solved. Summary of the Invention

[0008] To address this, the present invention provides an IoT-based sanitation operation scheduling method. Through a specially designed dual CEEMDAN model, it achieves efficient noise reduction and accurate decomposition of non-stationary time-series data on fuel consumption and load. A feature extraction model constructed using a multilayer perceptron, gating mechanism, and cross-modal attention mechanism enables effective fusion of multi-dimensional data. A mapping model using a long short-term memory network and a fully connected layer architecture corrects the vehicle's garbage load, achieving accurate fusion of multi-dimensional IoT data from garbage trucks. This allows for precise vehicle scheduling based on load, improving the operational efficiency of garbage trucks and reducing operating costs.

[0009] To achieve the above objectives, this invention proposes an Internet of Things-based sanitation operation scheduling method, comprising:

[0010] The vehicle load data stored in the sanitation operation scheduling platform is processed through the load CEEMDAN model to generate vehicle load characteristics, wherein the load CEEMDAN model determines the iteration stopping condition based on the time local weighted variance of the load data.

[0011] The vehicle IoT fuel consumption data stored in the sanitation operation dispatch platform is used to generate vehicle fuel consumption characteristics through the fuel consumption CEEMDAN model, wherein the fuel consumption CEEMDAN model calculates the fuel consumption channel residual signal based on the statistical values ​​of load data residual and fuel consumption data residual.

[0012] The vehicle fuel consumption characteristics, vehicle load characteristics, and IoT vehicle waste transfer operation data are used to generate comprehensive vehicle transport characteristics through a feature extraction model, wherein the feature extraction model is constructed based on a multilayer perceptron, a gating mechanism, and a cross-modal attention mechanism.

[0013] The comprehensive vehicle carrying characteristics are used to generate a corrected vehicle garbage load capacity through a mapping model based on a long short-term memory network and a fully connected layer architecture, and garbage transfer vehicle operations are scheduled according to the corrected vehicle garbage load capacity.

[0014] Furthermore, the process of generating vehicle fuel consumption characteristics using the CEEMDAN fuel consumption model includes:

[0015] The vehicle IoT fuel consumption data is processed through the fuel consumption CEEMDAN initial processing layer to generate fuel consumption data residuals;

[0016] The vehicle IoT load data is processed through the load CEEMDAN computing layer to generate load data residuals.

[0017] The coupling coefficient is calculated based on the covariance of the residuals of the fuel consumption data and the residuals of the load data, and the variance of the residuals of the fuel consumption data.

[0018] The fuel consumption channel noise is calculated based on the coupling coefficient, common noise signal, fuel consumption noise signal, and common noise amplitude.

[0019] The fuel consumption channel noise and fuel consumption data residuals are processed through the fuel consumption CEEMDAN iterative computation layer to generate the vehicle fuel consumption features.

[0020] The fuel consumption CEEMDAN model includes a fuel consumption CEEMDAN initial calculation layer and a fuel consumption CEEMDAN iterative calculation layer, and the load CEEMDAN model includes a load CEEMDAN calculation layer.

[0021] Furthermore, the process of generating the vehicle fuel consumption characteristics through the CEEMDAN iterative computation layer includes:

[0022] The first L2 norm is calculated for the high-frequency iterative IMF components generated by the CEEMDAN iterative computation layer, and the second L2 norm is calculated for the sum of the iterative IMF components. The ratio of the first L2 norm to the second L2 norm is used as the fuel consumption feature of the garbage loading action.

[0023] The third L2 norm is calculated for the mid-frequency iterative IMF component generated by the CEEMDAN iterative computation layer for fuel consumption, and the ratio of the third L2 norm to the second L2 norm is used as the fuel consumption feature of congested road sections.

[0024] The low-frequency iterative IMF component generated by the CEEMDAN iterative computation layer is used to calculate the fourth L2 norm, and the ratio of the fourth L2 norm to the second L2 norm is used as the overall fuel consumption feature of the garbage load.

[0025] The vehicle fuel consumption characteristics include fuel consumption characteristics of garbage loading actions, fuel consumption characteristics of congested road sections, and overall fuel consumption characteristics of garbage load. The iteration number of the high-frequency iterative IMF component, the medium-frequency iterative IMF component, and the low-frequency iterative IMF component increases sequentially.

[0026] Furthermore, the process of generating vehicle load characteristics using the load-bearing CEEMDAN model includes:

[0027] The load-bearing channel noise is calculated based on the aforementioned common noise signal and common noise amplitude.

[0028] The vehicle IoT load data and load channel noise are passed through the load CEEMDAN computation layer to iteratively calculate the load iterative IMF component.

[0029] The load iterative IMF components and vehicle IoT load data are passed through a local variance ratio stopping judgment layer to generate the vehicle load characteristics.

[0030] The load-bearing CEEMDAN model includes a load-bearing CEEMDAN computation layer and a local variance ratio stopping iteration judgment layer.

[0031] Furthermore, the process of stopping the iteration judgment layer to generate vehicle load characteristics by using the local variance ratio includes:

[0032] A first local weighted variance sequence is calculated based on the load iterative IMF components, and a second local weighted variance sequence is calculated based on the vehicle IoT load data;

[0033] Calculate the local variance ratio sequence based on the ratio of the first local weighted variance sequence and the second local weighted variance sequence;

[0034] The local instantaneous oscillation frequency is calculated based on the time interval between two adjacent zero-crossing moments of the load-iteration IMF component.

[0035] Based on the comparison result of the first median of the local variance ratio sequence and the stopping threshold, and the comparison result of the second median of the local instantaneous oscillation frequency and the loading action frequency range, it is determined whether to stop the iterative calculation of the load iterative IMF component. If stopped, the vehicle load characteristics are calculated based on the load iterative IMF component.

[0036] Furthermore, the process of generating comprehensive vehicle transport features through a feature extraction model includes:

[0037] The data on the waste transfer operations of the IoT vehicles are processed through a multilayer sensor to generate a transfer mapping vector.

[0038] The transfer mapping vector is passed through a gating mechanism to generate transfer gating weights, and the concatenated vector of vehicle fuel consumption features and vehicle load features is weighted based on the gating weights to generate weighted vehicle transfer volume features.

[0039] The transfer mapping vector and the weighted vehicle transfer volume features are used through a cross-modal attention mechanism to generate the comprehensive vehicle transport features.

[0040] Furthermore, the process of generating the integrated vehicle carrying features through a cross-modal attention mechanism includes:

[0041] The weighted vehicle transfer volume is passed through a standard convolutional layer to generate spatiotemporal features of vehicle transfer volume;

[0042] The spatiotemporal features of the vehicle transfer volume are processed through a flattening mapping layer to generate attention key vectors and attention value vectors;

[0043] The transport mapping vector is passed through an attention mapping layer to generate an attention query vector;

[0044] The attention key vector, attention value vector, and attention query vector are passed through the attention calculation layer to generate attention weighted features, and the concatenated vector of the attention weighted features and the spatiotemporal features of vehicle transfer volume is used as the comprehensive vehicle transport feature.

[0045] The cross-modal attention mechanism includes a standard convolutional layer, a flattening mapping layer, an attention mapping layer, and an attention computation layer.

[0046] Furthermore, the process of generating the transfer mapping vector through the multilayer perceptron includes:

[0047] The data from the IoT vehicle waste transfer operation is processed through a global average pooling layer to generate pooled features;

[0048] The pooled features are passed through a first convolutional layer to generate initial mapping features;

[0049] The initial mapping features are passed through a second convolutional layer to generate the transport mapping vector;

[0050] The multilayer perceptron includes a global average pooling layer, a first convolutional layer, and a second convolutional layer.

[0051] Furthermore, the process of generating a corrected vehicle garbage load capacity through a mapping model includes:

[0052] The integrated vehicle transport features are flattened to generate an integrated vehicle transport feature vector.

[0053] The integrated vehicle carrying feature vector is passed through a long short-term memory network to generate a total temporal hidden state;

[0054] The total temporal hidden state is passed through a fully connected layer to generate the corrected vehicle garbage load capacity.

[0055] Furthermore, the process of scheduling garbage transfer truck operations based on the corrected garbage load capacity of the vehicles includes:

[0056] The garbage load balance of the fleet is calculated based on the corrected vehicle garbage load weight, and the operation scheduling of garbage transfer vehicles is adjusted according to the garbage load balance of the fleet.

[0057] Compared with the prior art, the beneficial effects of the present invention are that, through the CEEMDAN data processing mechanism, multimodal deep feature fusion model and precise load correction, the present invention achieves efficient calibration and intelligent scheduling of garbage load of sanitation vehicles, avoids the problems of poor accuracy in non-stationary data processing and poor multi-source data fusion capability of scheduling methods, improves the operating efficiency of garbage transfer vehicles and reduces operating costs.

[0058] In particular, this invention strengthens the correlation and synergy between fuel consumption data and load data through the CEEMDAN fuel consumption model, obtains the overall fuel consumption characteristics of garbage load that reflect the core impact of load, realizes deep collaborative verification of dual data residuals, improves the reliability and consistency of the two types of core data, enhances the pertinence and characterization ability of fuel consumption characteristics, and thus improves the accuracy of vehicle garbage load correction through the Internet of Things.

[0059] In particular, the feature extraction model built upon multilayer perceptron, gating mechanism, and cross-modal attention mechanism of this invention achieves deep coupling and efficient fusion of multi-source heterogeneous data through a hierarchical and progressive processing flow, significantly enhancing feature representation capabilities. The gating mechanism generates transfer gating weights based on transfer mapping vectors, and performs weighted calculations on the concatenated vectors of vehicle fuel consumption features and vehicle load features, accurately enhancing core features related to load capacity. Deep fusion is achieved through the cross-modal attention mechanism, realizing cross-modal collaboration of operational data, fuel consumption features, and load features, providing comprehensive feature support for subsequent load capacity correction, thereby improving the accuracy of vehicle waste load capacity correction via the Internet of Things. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating the IoT-based sanitation operation scheduling method according to an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of the fuel consumption CEEMDAN model of the sanitation operation scheduling method based on the Internet of Things in this embodiment of the invention.

[0062] Figure 3 This is a schematic diagram of the load-bearing CEEMDAN model of the IoT-based sanitation operation scheduling method according to an embodiment of the present invention.

[0063] Figure 4 This is a schematic diagram of the feature extraction model of the Internet of Things-based sanitation operation scheduling method according to an embodiment of the present invention. Detailed Implementation

[0064] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0065] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0066] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0067] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0068] like Figures 1 to 4 As shown, this invention provides an IoT-based sanitation operation scheduling method. Through a specially designed dual CEEMDAN model, it achieves efficient noise reduction and accurate decomposition of non-stationary time-series data on fuel consumption and load. A feature extraction model constructed using a multilayer perceptron, gating mechanism, and cross-modal attention mechanism enables effective fusion of multi-dimensional data. A mapping model using a long short-term memory network and a fully connected layer architecture corrects the vehicle's garbage load, achieving accurate fusion of multi-dimensional IoT data from garbage trucks. This precise correction of vehicle load during scheduling improves the operational efficiency of garbage trucks and reduces operating costs.

[0069] like Figure 1 As shown, this embodiment proposes an IoT-based sanitation operation scheduling method, including:

[0070] The vehicle load data stored in the sanitation operation scheduling platform is processed through the load CEEMDAN model to generate vehicle load characteristics, wherein the load CEEMDAN model determines the iteration stopping condition based on the time local weighted variance of the load data.

[0071] The vehicle IoT fuel consumption data stored in the sanitation operation dispatch platform is used to generate vehicle fuel consumption characteristics through the fuel consumption CEEMDAN model, wherein the fuel consumption CEEMDAN model calculates the fuel consumption channel residual signal based on the statistical values ​​of load data residual and fuel consumption data residual.

[0072] The vehicle fuel consumption characteristics, vehicle load characteristics, and IoT vehicle waste transfer operation data are used to generate comprehensive vehicle transport characteristics through a feature extraction model, wherein the feature extraction model is constructed based on a multilayer perceptron, a gating mechanism, and a cross-modal attention mechanism.

[0073] The comprehensive vehicle carrying characteristics are used to generate a corrected vehicle garbage load capacity through a mapping model based on a long short-term memory network and a fully connected layer architecture, and garbage transfer vehicle operations are scheduled according to the corrected vehicle garbage load capacity.

[0074] Specifically, the vehicle IoT fuel consumption data and vehicle IoT load data are obtained through the on-board diagnostic (OBD) system on the garbage truck. The on-board diagnostic (OBD) system communicates with the cloud-based sanitation operation dispatch platform through a 4G or 5G data transmission module.

[0075] Specifically, this embodiment is preferably used for garbage transfer vehicles to transport garbage from communities, streets, and parks to garbage treatment stations. In this operation segment, the garbage transfer vehicles frequently start and stop, and the road bumps cause the garbage to frequently impact the load sensor. As a result, the vehicle IoT load data collected by the load sensor cannot accurately reflect the garbage load. However, based on the correlation characteristics of historical load and fuel consumption, the approximate load state can be inferred from the fuel consumption data, thereby obtaining a more accurate actual load of the garbage transfer vehicle. It can also correct the difference in vehicle IoT load data for the same weight of high-density kitchen waste and low-density waste paper.

[0076] Specifically, compared to the EMD model which is suitable for stationary nonlinear signals and the EEMD model which is suitable for non-stationary signals with a lot of noise, the CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) model in this embodiment is adapted to the characteristics of nonlinearity, non-stationarity, and multi-frequency coupling. It is used to decompose the original time series into multiple stationary IMF components, among which the high-frequency IMF components are used to reflect the impact of sudden fluctuations in garbage loading at the garbage collection point on the vehicle's carrying capacity.

[0077] like Figure 2 As shown, the process of generating vehicle fuel consumption characteristics using the CEEMDAN fuel consumption model further includes:

[0078] The vehicle IoT fuel consumption data is processed through the fuel consumption CEEMDAN initial processing layer to generate fuel consumption data residuals;

[0079] The vehicle IoT load data is processed through the load CEEMDAN computing layer to generate load data residuals.

[0080] The coupling coefficient is calculated based on the covariance of the residuals of the fuel consumption data and the residuals of the load data, and the variance of the residuals of the fuel consumption data.

[0081] The fuel consumption channel noise is calculated based on the coupling coefficient, common noise signal, fuel consumption noise signal, and common noise amplitude.

[0082] The fuel consumption channel noise and fuel consumption data residuals are processed through the fuel consumption CEEMDAN iterative computation layer to generate the vehicle fuel consumption features.

[0083] The fuel consumption CEEMDAN model includes a fuel consumption CEEMDAN initial calculation layer and a fuel consumption CEEMDAN iterative calculation layer, and the load CEEMDAN model includes a load CEEMDAN calculation layer.

[0084] Specifically, the process of the initial computation layer of CEEMDAN can be represented as follows:

[0085]

[0086]

[0087]

[0088]

[0089] In the formula, This represents the vehicle's IoT fuel consumption data at time t. This represents the initial residual of fuel consumption at time t. This represents the initial noise amplitude, preferably 0.2. This represents an independent common noise signal, which is random white noise. This represents the first-order intrinsic mode function (IMF) of the EMD decomposition. The first-order IMF corresponds to the highest frequency component of the signal, which corresponds to the sudden fluctuations in the load during garbage truck loading and unloading. This represents the first-order screening operator for EMD decomposition. Represents the first-order IMF component. This means that by taking the arithmetic mean of the first-order intrinsic mode functions of N noisy signals, the random effects of white noise are canceled out, resulting in a stable IMF without noise pollution. This represents the residual signal after the first-order decomposition, i.e., the residual of the fuel consumption data.

[0090] Specifically, the iterative calculation process of the fuel consumption channel noise and the fuel consumption CEEMDAN iterative operation layer can be represented as follows:

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097] In the formula, This represents the coupling coefficient for the t-th time window. This indicates the calculation of covariance. This indicates variance calculation. Let K represent the load data residual and fuel consumption data residual from the previous iteration, respectively, where the iteration number k = 2, 3, 4, until the iteration stopping condition is met. The value is taken from the residual fuel consumption data output from the initial CEEMDAN calculation layer, or the iterative output from the CEEMDAN iterative calculation layer. This represents the minimum value to prevent the denominator from being zero, preferably 1e-8. It dynamically describes the rate of change in fuel consumption caused by a change in unit load under the current operating conditions. This indicates the noise level in the fuel consumption channel. Represents the amplitude of common noise, preferably / k, Let be the initial noise amplitude, preferably 0.2, and k represent the iteration number, so that the common noise amplitude decreases as the decomposition order increases. The common noise signal is an independent Gaussian white noise sequence. The common noise amplitude and the common noise signal are common to both the fuel consumption signal and the load signal, and are used to introduce a consistent disturbance pattern. This represents the fuel consumption noise signal, which is independent of the common noise signal. It represents random disturbances in the fuel consumption signal that are not directly affected by load changes, such as engine fluctuations and sensor thermal noise, and is used to maintain the completeness of the decomposition. Let i represent the i-th signal to be decomposed from a total of N groups. Adaptive fuel consumption channel noise is injected into the previous residual for signal decomposition. This represents the first-order screening operator for EMD decomposition. This represents the candidate IMF component in the k-th iteration. This represents the IMF component in the k-th iteration. The residual signal after the k-th iteration decomposition is used as the residual signal of the load data of the previous iteration in the next iteration. Iterative calculation is performed using the above-mentioned coupling coefficient and other calculation formulas until the residual signal of the load data of the previous iteration is a monotonic function or a constant, and it is impossible to decompose it into components that satisfy the IMF condition, or the residual signal of the load data of the previous iteration is less than 5% of the signal amplitude of the original vehicle IoT fuel consumption data, that is, its contribution to signal reconstruction is negligible. At this time, the iteration stops, and the linear fitting slope of the residual signal of the load data of the previous iteration is used as a component of the vehicle fuel consumption feature.

[0098] Furthermore, the process of generating the vehicle fuel consumption characteristics through the CEEMDAN iterative computation layer includes:

[0099] The first L2 norm is calculated for the high-frequency iterative IMF components generated by the CEEMDAN iterative computation layer, and the second L2 norm is calculated for the sum of the iterative IMF components. The ratio of the first L2 norm to the second L2 norm is used as the fuel consumption feature of the garbage loading action.

[0100] The third L2 norm is calculated for the mid-frequency iterative IMF component generated by the CEEMDAN iterative computation layer for fuel consumption, and the ratio of the third L2 norm to the second L2 norm is used as the fuel consumption feature of congested road sections.

[0101] The low-frequency iterative IMF component generated by the CEEMDAN iterative computation layer is used to calculate the fourth L2 norm, and the ratio of the fourth L2 norm to the second L2 norm is used as the overall fuel consumption feature of the garbage load.

[0102] The vehicle fuel consumption characteristics include fuel consumption characteristics of garbage loading actions, fuel consumption characteristics of congested road sections, and overall fuel consumption characteristics of garbage load. The iteration number of the high-frequency iterative IMF component, the medium-frequency iterative IMF component, and the low-frequency iterative IMF component increases sequentially.

[0103] Specifically, the process of calculating the fuel consumption characteristics of garbage loading operations, congested road sections, and overall garbage load can be expressed as follows:

[0104]

[0105] In the formula, This represents the ratio of the L2 norm of the iterative IMF component in the k-th iteration to the L2 norm of the sum of the iterative IMF components, i.e., when k takes the values ​​1 and 2. To correspond to the high-frequency iterative IMF components that represent the brief impact caused by a single loading action, when k takes values ​​of 3 and 4, For the mid-frequency iterative IMF component corresponding to the stop-and-go pattern in congested road sections, when k takes the values ​​K-1 and K, To correspond to the low-frequency iterative IMF components that reflect the overall trend of load, This represents the IMF component in the k-th iteration. Let represent the IMF component in the j-th iteration.

[0106] like Figure 3 As shown, the process of generating vehicle load characteristics using the load-bearing CEEMDAN model further includes:

[0107] The load-bearing channel noise is calculated based on the aforementioned common noise signal and common noise amplitude.

[0108] The vehicle IoT load data and load channel noise are passed through the load CEEMDAN computation layer to iteratively calculate the load iterative IMF component.

[0109] The load iterative IMF components and vehicle IoT load data are passed through a local variance ratio stopping judgment layer to generate the vehicle load characteristics.

[0110] The load-bearing CEEMDAN model includes a load-bearing CEEMDAN computation layer and a local variance ratio stopping iteration judgment layer.

[0111] Specifically, the iterative calculation process of the load channel noise and the load CEEMDAN operation layer can be expressed as follows:

[0112]

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121] In the formula, This represents vehicle IoT load data. This represents the initial residual of the load at time t. This represents the initial noise amplitude, preferably 0.2. This represents a common noise signal, which is random white noise. This represents the first-order eigenmode function of the load in the EMD decomposition. This represents the first-order screening operator for EMD decomposition. Represents the first-order IMF component. This means that the arithmetic mean of the first-order intrinsic mode functions of the loads of N noisy signals is taken to cancel out the random effects of white noise, resulting in a stable IMF without noise pollution. This represents the residual signal after the first-order decomposition, i.e., the load data residual. Indicates the noise level in the load-bearing passageway. This represents the load residual signal after the (k-1)th iteration decomposition, from which the iterative calculation process begins. This represents the k-th iteration candidate IMF component of the load in the EMD decomposition. This represents the load iteration IMF component of the k-th iteration. This represents the load residual signal after the kth iteration decomposition. The local variance ratio is used to determine whether to end the iteration. If the iteration is not ended, it is used as the new load residual signal after the (k-1)th iteration decomposition and recalculated using the above formula.

[0122] Furthermore, the process of stopping the iteration judgment layer to generate vehicle load characteristics by using the local variance ratio includes:

[0123] A first local weighted variance sequence is calculated based on the load iterative IMF components, and a second local weighted variance sequence is calculated based on the vehicle IoT load data;

[0124] Calculate the local variance ratio sequence based on the ratio of the first local weighted variance sequence and the second local weighted variance sequence;

[0125] The local instantaneous oscillation frequency is calculated based on the time interval between two adjacent zero-crossing moments of the load-iteration IMF component.

[0126] Based on the comparison result of the first median of the local variance ratio sequence and the stopping threshold, and the comparison result of the second median of the local instantaneous oscillation frequency and the loading action frequency range, it is determined whether to stop the iterative calculation of the load iterative IMF component. If stopped, the vehicle load characteristics are calculated based on the load iterative IMF component.

[0127] Specifically, the local variance ratio stopping iteration judgment layer can be expressed as:

[0128]

[0129]

[0130]

[0131] In the formula, This represents the local variance ratio sequence at time t. , Let represent the first local weighted variance sequence and the second local weighted variance sequence, respectively. Here, t represents the time of data collection for the vehicle IoT load data, T represents the duration of the local time window (set to be slightly longer than the duration of a typical loading or unloading operation; for example, if the average loading time is 3 minutes and the sampling rate is 1Hz, then T is set to 180), and w represents the vehicle IoT load data. This represents the minimum value to prevent the denominator from being zero, preferably 1e-8. This represents the load iteration IMF component of the k-th iteration. This describes the calculation process of the first locally weighted variance sequence, where... For time variables, The weighting function is preferably... This further emphasizes the contribution of the window center. This represents the average value of multiple load iterations of the IMF component within a local time window. This represents the average value of multiple vehicle IoT load data points within a local time window. This represents the first median of the local variance ratio series. This indicates the calculation of the median. This represents the frequency of the j-th local instantaneous oscillation. This represents the difference in timestamps between the j-th and (j+1)-th adjacent zero-crossing points of the load-bearing iterative IMF component in the k-th iteration. The zero-crossing point of the load-bearing iterative IMF component, i.e., the moment the signal value crosses the zero horizontal line, marks the beginning or end of an oscillation cycle. At that time, the load iteration IMF component of this iteration is used as the vehicle load feature, where The preferred value is 0.2. This represents the median of the local instantaneous oscillation frequency. This represents the frequency range of loading actions, specifically the frequency range from opening to closing the loading chamber during a single waste loading action. Therefore, the above formula can be used to determine, through an adaptive stopping criterion, whether the candidate components generated during the CEEMDAN screening process have iterated to the desired modal characteristics that represent a complete operation event.

[0132] like Figure 4 As shown, the process of generating comprehensive vehicle carrying features through the feature extraction model further includes:

[0133] The data on the waste transfer operations of the IoT vehicles are processed through a multilayer sensor to generate a transfer mapping vector.

[0134] The transfer mapping vector is passed through a gating mechanism to generate transfer gating weights, and the concatenated vector of vehicle fuel consumption features and vehicle load features is weighted based on the gating weights to generate weighted vehicle transfer volume features.

[0135] The transfer mapping vector and the weighted vehicle transfer volume features are used through a cross-modal attention mechanism to generate the comprehensive vehicle transport features.

[0136] Specifically, the IoT vehicle waste transfer operation data includes unique thermal codes for waste transfer points containing waste types, unique thermal codes for the weather conditions at the location of the waste transfer vehicle, and codes for waste transfer time.

[0137] Furthermore, the process of generating the integrated vehicle carrying features through a cross-modal attention mechanism includes:

[0138] The weighted vehicle transfer volume is passed through a standard convolutional layer to generate spatiotemporal features of vehicle transfer volume;

[0139] The spatiotemporal features of the vehicle transfer volume are processed through a flattening mapping layer to generate attention key vectors and attention value vectors;

[0140] The transport mapping vector is passed through an attention mapping layer to generate an attention query vector;

[0141] The attention key vector, attention value vector, and attention query vector are passed through the attention calculation layer to generate attention weighted features, and the concatenated vector of the attention weighted features and the spatiotemporal features of vehicle transfer volume is used as the comprehensive vehicle transport feature.

[0142] The cross-modal attention mechanism includes a standard convolutional layer, a flattening mapping layer, an attention mapping layer, and an attention computation layer.

[0143] Furthermore, the process of generating the transfer mapping vector through the multilayer perceptron includes:

[0144] The data from the IoT vehicle waste transfer operation is processed through a global average pooling layer to generate pooled features;

[0145] The pooled features are passed through a first convolutional layer to generate initial mapping features;

[0146] The initial mapping features are passed through a second convolutional layer to generate the transport mapping vector;

[0147] The multilayer perceptron includes a global average pooling layer, a first convolutional layer, and a second convolutional layer.

[0148] Specifically, the feature extraction model can be represented as:

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156]

[0157] In the formula, This represents data on waste transfer operations by IoT vehicles. This indicates a global average pooling layer. , , These represent the pooling features, initial mapping features, and transport mapping vectors, respectively. express Activation function , , , Both represent the weight vector and bias vector of the convolution operation. Indicates the transfer gate weight, This represents the SIGMOD activation function. , Represents the weight vector and bias vector of the gating mechanism. , These represent the concatenated vectors of weighted vehicle transport volume characteristics, vehicle fuel consumption characteristics, and vehicle load characteristics, respectively. This represents multiplication along the channel dimension; therefore, the weighted vehicle transfer volume features tell the model which features should be emphasized or ignored under current weather and road conditions. This indicates the spatiotemporal characteristics of vehicle transfer volume. express Activation function Indicates batch normalization. , This represents the weight vector and bias vector of a standard convolutional layer, where Q, K, and V represent the attention query vector, attention key vector, and attention value vector, respectively. , , Both represent the mapping convolution weight matrix. Indicates the flattening operation. This represents the attention-weighted feature weights. express function, This represents the dimension of the attention query vector. This represents attention-weighted features. This indicates a feature concatenation operation. It indicates the comprehensive characteristics of vehicle transportation.

[0158] Furthermore, the process of generating a corrected vehicle garbage load capacity through a mapping model includes:

[0159] The integrated vehicle transport features are flattened to generate an integrated vehicle transport feature vector.

[0160] The integrated vehicle carrying feature vector is passed through a long short-term memory network to generate a total temporal hidden state;

[0161] The total temporal hidden state is passed through a fully connected layer to generate the corrected vehicle garbage load capacity.

[0162] Specifically, the mapping model can be represented as:

[0163]

[0164]

[0165]

[0166]

[0167]

[0168]

[0169]

[0170] In the formula, This represents the comprehensive vehicle carrying feature vector. This indicates a flattening operation, which involves flattening the vehicle in a spatial dimension, treating each spatial location as an element of a comprehensive vehicle carrying characteristic vector. , , , , , The sequence represents the forgetting gate, input gate, output gate, candidate memory, cell state update, and hidden state output of the Long Short-Term Memory (LSTM) network. This represents the SIGMOD activation function, where W and b in the formula represent the learnable weight matrix and bias vector, respectively. express Activation function This represents the total hidden temporal state across all time steps. express Activation function This indicates the corrected garbage load capacity of the vehicle.

[0171] Specifically, in this embodiment, the loss function of the feature extraction model and the mapping model is preferably a weighted combination of the mean absolute error loss function and the cross-entropy loss function, thereby simultaneously optimizing the load prediction accuracy and the operation event recognition capability.

[0172] Specifically, the kitchen waste transfer operation data of a municipal sanitation group from June to December 2023, including complete IoT records of 47 transfer vehicles and 3,856 collection points, were used to calculate the waste load using the following models: the traditional statistical model (benchmark 1), the pure LSTM prediction model (benchmark 2), the CNN-LSTM hybrid model without CEEMDAN preprocessing (benchmark 3), and the load CEEMDAN model, fuel consumption CEEMDAN model, feature extraction model, and mapping model of this embodiment. The mean absolute percentage error (MAPE) and root mean square error (RMSE) (kg) between the waste load and the actual waste load when the vehicle is stationary were evaluated. The results were as follows: benchmark 1: 8.7, 183.5; benchmark 2: 6.2, 126.8; benchmark 3: 4.8, 102.3; and this embodiment: 3.4, 78.5. As can be seen, the feature extraction model and mapping model in this embodiment have lower errors compared to other models. The overall CEEMDAN-CNN-LSTM model with load CEEMDAN model and fuel consumption CEEMDAN model in this embodiment has further reduced errors compared to the model without CEEMDAN model.

[0173] Furthermore, the process of scheduling garbage transfer truck operations based on the corrected garbage load capacity of the vehicles includes:

[0174] The fleet's waste load balance is calculated based on the corrected vehicle waste load capacity, and the waste transfer vehicle operation scheduling is adjusted according to the fleet's waste load balance. The fleet's waste load balance is... ,in This represents the corrected garbage load capacity of the i-th vehicle. This indicates the average garbage load capacity of the fleet. This indicates the balance of the fleet's waste load.

[0175] Specifically, the fleet's waste load balance can be used as an objective term in the waste transfer vehicle scheduling decision function, thereby achieving comprehensive waste transfer vehicle operation scheduling. Alternatively, based on a comparison of the fleet's waste load balance with a threshold, it can assess whether waste transfer vehicles in the fleet are either "star" vehicles about to be overloaded or underutilized idle vehicles. When assigning new tasks to "star" vehicles about to be overloaded, the system becomes more reluctant to allocate new tasks, potentially assigning the next high-load collection point, which is closer to the "star" vehicle, to an idle vehicle that is slightly further away but less busy. In the short term, this may seem to increase the driving distance of idle vehicles, but it avoids the subsequent route interruptions and wasted empty runs caused by "star" vehicles returning to the depot prematurely fully loaded.

[0176] In this embodiment, the efficient calibration and intelligent scheduling of garbage load capacity of sanitation vehicles are achieved through the CEEMDAN data processing mechanism, multimodal deep feature fusion model, and precise load capacity correction. This avoids the problems of poor accuracy in non-stationary data processing and poor multi-source data fusion capabilities of scheduling methods, thereby improving the operational efficiency of garbage transfer vehicles and reducing operating costs. The CEEMDAN fuel consumption model strengthens the correlation and synergy between fuel consumption data and load capacity data, obtaining overall fuel consumption characteristics of garbage load capacity that reflect the core impact of load. Deep collaborative verification of the residuals of the two data sets is achieved, improving the reliability and consistency of the two types of core data, enhancing the specificity and characterization ability of fuel consumption features, and thus improving the accuracy of vehicle garbage load capacity correction via the Internet of Things. The feature extraction model, built upon multilayer perceptron, gating mechanism, and cross-modal attention mechanism, achieves deep coupling and efficient fusion of multi-source heterogeneous data through a hierarchical processing flow, significantly enhancing feature representation capabilities. The gating mechanism generates transfer gating weights based on the transfer mapping vector, and performs weighted calculations on the concatenated vector of vehicle fuel consumption features and vehicle load features, accurately strengthening core features related to load capacity. The cross-modal attention mechanism achieves deep fusion, realizing cross-modal collaboration of operational data, fuel consumption features, and load features, providing comprehensive feature support for subsequent load capacity correction, thereby improving the accuracy of vehicle waste load capacity correction via the Internet of Things.

[0177] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0178] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0179] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A sanitation operation scheduling method based on the Internet of Things, characterized in that, include: The vehicle load data stored in the sanitation operation scheduling platform is processed through the load CEEMDAN model to generate vehicle load characteristics, wherein the load CEEMDAN model determines the iteration stopping condition based on the time local weighted variance of the load data. The vehicle IoT fuel consumption data stored in the sanitation operation dispatch platform is used to generate vehicle fuel consumption characteristics through the fuel consumption CEEMDAN model, wherein the fuel consumption CEEMDAN model calculates the fuel consumption channel residual signal based on the statistical values ​​of load data residual and fuel consumption data residual. The vehicle fuel consumption characteristics, vehicle load characteristics, and IoT vehicle waste transfer operation data are used to generate comprehensive vehicle transport characteristics through a feature extraction model, wherein the feature extraction model is constructed based on a multilayer perceptron, a gating mechanism, and a cross-modal attention mechanism. The comprehensive vehicle carrying characteristics are used to generate a corrected vehicle garbage load capacity through a mapping model based on a long short-term memory network and a fully connected layer architecture, and garbage transfer vehicle operation scheduling is carried out based on the corrected vehicle garbage load capacity. The process of generating vehicle fuel consumption characteristics using the CEEMDAN fuel consumption model includes: The vehicle IoT fuel consumption data is processed through the fuel consumption CEEMDAN initial processing layer to generate fuel consumption data residuals; The vehicle IoT load data is processed through the load CEEMDAN computing layer to generate load data residuals. The coupling coefficient is calculated based on the covariance of the residuals of the fuel consumption data and the residuals of the load data, and the variance of the residuals of the fuel consumption data. The fuel consumption channel noise is calculated based on the coupling coefficient, common noise signal, fuel consumption noise signal, and common noise amplitude. The fuel consumption channel noise and fuel consumption data residuals are processed through the fuel consumption CEEMDAN iterative computation layer to generate the vehicle fuel consumption features. The fuel consumption CEEMDAN model includes a fuel consumption CEEMDAN initial calculation layer and a fuel consumption CEEMDAN iterative calculation layer, and the load CEEMDAN model includes a load CEEMDAN calculation layer. The process of generating vehicle load characteristics using the CEEMDAN load model includes: The load-bearing channel noise is calculated based on the aforementioned common noise signal and common noise amplitude. The vehicle IoT load data and load channel noise are passed through the load CEEMDAN computation layer to iteratively calculate the load iterative IMF component. The load iterative IMF components and vehicle IoT load data are passed through a local variance ratio stopping judgment layer to generate the vehicle load characteristics. The load-bearing CEEMDAN model includes a load-bearing CEEMDAN computation layer and a local variance ratio stopping iteration judgment layer.

2. The sanitation operation scheduling method based on the Internet of Things according to claim 1, characterized in that, The process of generating the vehicle fuel consumption characteristics through the CEEMDAN iterative computation layer includes: The first L2 norm is calculated for the high-frequency iterative IMF components generated by the CEEMDAN iterative computation layer, and the second L2 norm is calculated for the sum of the iterative IMF components. The ratio of the first L2 norm to the second L2 norm is used as the fuel consumption feature of the garbage loading action. The third L2 norm is calculated for the mid-frequency iterative IMF component generated by the CEEMDAN iterative computation layer for fuel consumption, and the ratio of the third L2 norm to the second L2 norm is used as the fuel consumption feature of congested road sections. The low-frequency iterative IMF component generated by the CEEMDAN iterative computation layer is used to calculate the fourth L2 norm, and the ratio of the fourth L2 norm to the second L2 norm is used as the overall fuel consumption feature of the garbage load. The vehicle fuel consumption characteristics include fuel consumption characteristics of garbage loading actions, fuel consumption characteristics of congested road sections, and overall fuel consumption characteristics of garbage load. The iteration number of the high-frequency iterative IMF component, the medium-frequency iterative IMF component, and the low-frequency iterative IMF component increases sequentially.

3. The sanitation operation scheduling method based on the Internet of Things according to claim 1, characterized in that, The process of generating vehicle load characteristics by stopping iterations based on local variance ratio includes: A first local weighted variance sequence is calculated based on the load iterative IMF components, and a second local weighted variance sequence is calculated based on the vehicle IoT load data; Calculate the local variance ratio sequence based on the ratio of the first local weighted variance sequence and the second local weighted variance sequence; The local instantaneous oscillation frequency is calculated based on the time interval between two adjacent zero-crossing moments of the load-iteration IMF component. Based on the comparison result of the first median of the local variance ratio sequence and the stopping threshold, and the comparison result of the second median of the local instantaneous oscillation frequency and the loading action frequency range, it is determined whether to stop the iterative calculation of the load iterative IMF component. If stopped, the vehicle load characteristics are calculated based on the load iterative IMF component.

4. The sanitation operation scheduling method based on the Internet of Things according to claim 1, characterized in that, The process of generating comprehensive vehicle carrying characteristics through a feature extraction model includes: The data on the waste transfer operations of the IoT vehicles are processed through a multilayer sensor to generate a transfer mapping vector. The transfer mapping vector is passed through a gating mechanism to generate transfer gating weights, and the concatenated vector of vehicle fuel consumption features and vehicle load features is weighted based on the gating weights to generate weighted vehicle transfer volume features. The transfer mapping vector and the weighted vehicle transfer volume features are used through a cross-modal attention mechanism to generate the comprehensive vehicle transport features.

5. The sanitation operation scheduling method based on the Internet of Things according to claim 4, characterized in that, The process of generating the integrated vehicle carrying characteristics through a cross-modal attention mechanism includes: The weighted vehicle transfer volume is passed through a standard convolutional layer to generate spatiotemporal features of vehicle transfer volume; The spatiotemporal features of the vehicle transfer volume are processed through a flattening mapping layer to generate attention key vectors and attention value vectors; The transport mapping vector is passed through an attention mapping layer to generate an attention query vector; The attention key vector, attention value vector, and attention query vector are passed through the attention calculation layer to generate attention weighted features, and the concatenated vector of the attention weighted features and the spatiotemporal features of vehicle transfer volume is used as the comprehensive vehicle transport feature. The cross-modal attention mechanism includes a standard convolutional layer, a flattening mapping layer, an attention mapping layer, and an attention computation layer.

6. The sanitation operation scheduling method based on the Internet of Things according to claim 4, characterized in that, The process of generating transport mapping vectors using a multilayer perceptron includes: The data from the IoT vehicle waste transfer operation is processed through a global average pooling layer to generate pooled features; The pooled features are passed through a first convolutional layer to generate initial mapping features; The initial mapping features are passed through a second convolutional layer to generate the transport mapping vector; The multilayer perceptron includes a global average pooling layer, a first convolutional layer, and a second convolutional layer.

7. The sanitation operation scheduling method based on the Internet of Things according to claim 1, characterized in that, The process of generating a corrected vehicle garbage load capacity using a mapping model includes: The integrated vehicle transport features are flattened to generate an integrated vehicle transport feature vector. The integrated vehicle carrying feature vector is passed through a long short-term memory network to generate a total temporal hidden state; The total temporal hidden state is passed through a fully connected layer to generate the corrected vehicle garbage load capacity.

8. The sanitation operation scheduling method based on the Internet of Things according to any one of claims 1 to 7, characterized in that, The process of scheduling garbage transfer truck operations based on the corrected garbage load capacity of the vehicles includes: The garbage load balance of the fleet is calculated based on the corrected vehicle garbage load weight, and the operation scheduling of garbage transfer vehicles is adjusted according to the garbage load balance of the fleet.

Images